Add a fail-closed, fresh-process native trace workflow for the Offline Seasons
fixture-to-match-team boundary. The supervisor ignores UMU's short-lived
FIFA17.exe process, requires CardsDLL, verifies TracerPid and hardware arming,
rejects pre-existing records, structurally locates fixtures/final records, and
detaches cleanly after capture.
The four payloads reproduce the measured chain without client writes:
FUN_1800fc500
-> actual season vector, fixture index 0 / team 73
-> temporary [73,130000] pair (not the final record)
CardsDLL service -> engine 0x147c652ce
-> live final +0x14 writes at the 0x45c side stride
-> correct [73,130000], then local overwrite [130000,130000]
engine wrapper 0x147ce47e0
<- CardsDLL 0x180031861
<- CardsGameSetupAdapter local `teams` query result already 130000
All execute breakpoints and watchpoints are hardware-only. /proc/PID/mem is
opened rb. No INT3, write_memory, patch, game input, server behavior, or Rust
code. Locator uses zero-based --fixture-index (the live selector is 0 when
season/user.round is 1) and never filters on transient +0x18 handles.
Read-only probes for resolving FIFA17 code paths against a running client
without Ghidra, per the live-disassembly method (/proc/<pid>/mem + objdump).
All open /proc/<pid>/mem 'rb' only.
ldis.py image-VA disassembler/hexdump for CardsDLL and FIFA17.exe;
recomputes the module base from the NAMED PE-header mapping
every run, because Wine maps PE sections anonymously and the
mapping that merely CONTAINS an address is not the module.
xref.py references to an image VA: call/jmp rel32, rip-relative lea,
and absolute pointer slots. An absolute-only hit means the
function is virtual and reachable solely via its vtable.
immstore.py immediate stores (C7 /0) of a constant to a struct offset.
Only an immediate store can INTRODUCE a constant; a register
store merely propagates one. Zero hits is a real result: it
proves the constant arrives from a call, not a literal.
classify_calls.py splits call sites of a constant-returning stub into STORE
(can assign) vs compare (predicate). Turned 81 call sites of
the 130000 provider into 17 assignments.
vtab.py dumps a vtable as image VAs and looks for sibling vtables
holding a different function in the same slot, which is how
a type/mode dispatch shows up.
scan_mt.py match-team records by the invariant header (11,7,0,0,76).
Never filters on +0x18: that word is a per-session handle
(-1 on 2026-08-24, 0x54001/0x54000 on 2026-08-25) and
filtering on it previously produced a false negative.
Workflow note: dump .text once and cache the objdump output, then query the
cached listing; a full CardsDLL .text linear disassembly is ~563k lines and
re-disassembling per question is wasteful.
Promotes the throwaway probe used to close the manager cold-load milestone into
fifa17-recon/tools. Read-only (/proc/<pid>/mem opened 'rb', never 'r+b'), pid
optional and overridable, controls overridable via --control WIRE:RESOURCE.
Fail-closed: absent player positive controls exit 3 (INCONCLUSIVE, squad not
loaded) rather than 0, so 'no manager found' can never be reported from a
session that never loaded a squad. Distinguishes real item records from
incidental integer matches by requiring resourceId 0x20 bytes before the wire
id, the layout the player controls exhibit.
Documents the manager wire control, the resourceId control (the actual
verdict), the player positive controls, and what counts as a resident hit.
An owned manager assigned in Core was present everywhere on the server -- in
/club/manager, in club?type=staff, and in userMassInfo -- but the squad UI
showed no manager after a cold client load.
The squad parser FUN_18013d1f0 reaches the item parser FUN_18013fe00 by two
different routes:
players: atom 568 -> per-element atoms 355 `index`, 363 `itemData`,
378 `kitNumber`; the 363 arm at 0x18013d8d9 calls the item parser
on the NESTED itemData object.
manager: atom 424 -> array loop at 0x18013da29 calls that same item parser
DIRECTLY on the array ELEMENT, into squad+0xC0. No `itemData` step.
So a manager element IS an item. We were nesting the fields one level deeper,
so the parser read only the two keys that happen to be item atoms -- `id` and
`dream` -- and left everything else at its default. Measured on a cold client,
the manager record existed at squad+0xC0 with the correct id and resourceId 0,
while sibling players in the same response carried theirs. resourceId is the
merge key compared RAW against carddbid, so 0 resolves no manager: no name, no
rating, no art, empty slot.
The client's own save corroborates the shape: it PUTs
`"manager":[{"id":...,"dream":false}]` -- flat, and both keys are item atoms.
Flatten the element to the item plus `dream`. Cold-load proven on staging: the
manager record now carries resourceId 1000509 in the same layout as its player
siblings (83906881, 84053575) in the same array region, and the operator
confirms a manager is assigned in the squad management screen.
Two earlier shapes are now both explained and covered by tests: `{id, dream}`
carries no merge key, and `{id, itemData, dream}` hides it from this path.
FIFA 17 sends two different operations to `PUT …/squad/<id>` and distinguishes
them only by body shape. Across 73 captured squad PUTs in five captures there
are exactly two:
* 68x with `players` -- a full replacement (also carrying squadName,
formation, squadType, manager, chemistry/rating, and redundantly
captain/kicktakers);
* 5x without `players` -- `{id, custom, captain, kicktakers}`, emitted by the
captain/kick-taker screen.
`players` has `#[serde(default)]`, so an absent key and an explicit `[]`
collapsed to the same empty vec and every partial update was handed to Core as
a replacement with zero slots. Core's empty-replacement guard refused it (400)
and the host reported 502, losing the user's captain/kick-taker change.
`classify_squad_put` now tests key PRESENCE on the raw JSON before
deserialising, so absence ("the squad was not part of this edit") stays
distinct from an explicit empty array ("replace with nothing"). An explicit
`"players": []` still classifies as a replacement and still meets the guard --
the patch path is not a way around it.
The patch path carries the contract correction: omitted `players`, `manager`
and actives mean UNCHANGED, never cleared. That is structural --
`CoreRolePatchRequest` has no field able to express them. The extension is
MERGED rather than overwritten, because the partial body carries only `custom`
and `kicktakers`; overwriting would drop every kit number in the squad.
`custom` IS taken from the patch, since the role screen writes per-slot values
into it and the two shapes genuinely differ there.
Also fixes the error mapping on this route: a Core 400 means the REQUEST was
invalid, so it is reported as 400, not as a 502 that blames the server and
hides a client error behind "upstream unavailable".
Tests use the real captured body and assert it takes the patch path
(`replace_squad` call count unchanged), that the manager survives, that kit
numbers survive the merge, that an explicit empty `players` still reaches the
replacement path, and that an unresolvable captain refuses the whole patch
rather than half-applying the kick-takers.
The numeric id in squad/<n> was being justified as "matching the oracle". That
justification does not survive inspection, and the code now says why.
FIFA 17 has genuine multi-squad semantics. The client's own shipped action
table has SelectSquadById, RetrieveSquad as an action DISTINCT from
LoadActiveSquad, CreateSquadWithName, RenameSquad, DeleteSquad, CopySquad,
indexed SQUAD_ID-%d list entries and FUT_MAX_NUM_SQUAD_REACHED. The base
template is `ut/%s/squad` with the id appended. The number identifies a squad.
The inherited behaviour came from a bare prefix regex in the Python oracle -
`re.compile(G + r"/squad")` calling squad_route(), which never reads the URL id
(GET returns current_squad(), PUT echoes the id from the BODY). The comments
around it show /squad/list and the draft routes had to be registered first
because that rule was swallowing them. It was expedient, not evidence-driven.
Collapsing the id is nonetheless SAFE today, and only for a specific reason:
we advertise exactly one squad. ACTIVE_SQUAD_WIRE_ID is a constant 0,
/squad/list returns a single-element array carrying it, and no
create/rename/delete/copy route exists, so the client can only echo back the id
we gave it. Every numeric path in retained captures is squad/0, all PUTs whose
body id also reads 0; no numeric GET has ever been recorded.
Behaviour is therefore UNCHANGED - no evidence justifies changing it, and
unknown-id semantics are deliberately not invented. What changes is that the
assumption is now explicit and enforced:
numeric_squad_routing_is_safe_only_while_one_squad_is_advertised pins the wire
id at 0 and /squad/list at one entry, and fails the moment a second squad
becomes addressable. Verified by simulating a second advertised squad.
Workspace 1252 passed (1251 + this test), 0 failed.
The retail client sends BOTH casings. Observed twice on staging, each time
inside a genuine client sequence:
16:56:56 route=squad-active 200
16:56:56 GET /ut/game/fifa17/usermassinfo -> passthrough -> 502
16:56:56 route=userMassInfo 200
classify matched the exact literal `userMassInfo`, so the lowercase request
fell through to the Python upstream. Today that is a harmless 502 because the
upstream is dead and the client immediately retries with the canonical casing -
but on a deployment with Python ALIVE that request would be ANSWERED there,
silently splitting authority away from Rust for a route Core owns. That is the
real defect, not the wasted round trip.
Fixed with the smallest possible alias: this one tail is matched
case-insensitively, the rest of the table stays exact since no other route has
ever shown a casing variant. Paths are NOT globally lowercased.
Tests cover the canonical casing, lowercase, uppercase, two adjacent tails that
must NOT be swept up by the alias (`usermassinfox`, `usermass`), and method
semantics (PUT/POST still passthrough). With the alias reverted the test fails.
The squad_actives default test was inserted between #[test] and the function
it belonged to, which stacked a duplicate attribute on the new test and left
sbc_post_commit_faults_require_all_staging_guards with none - silently
disabling it while the new test ran twice.
Caught by clippy (-D duplicate-macro-attributes, -D dead-code); the doubled
test name in the earlier run was the tell. Lib tests go 129 -> 131 with both
now executing.
Serving the club's active home and away kit is normal FIFA 17 behaviour, not
an experiment: it is what lets the client make the kits resident and render
the pre-match selector. A correct deployment should not have to opt in, so the
default is now ON and the environment variable survives only as a diagnostic
off switch (OPENFUT_FIFA17_SQUAD_ACTIVES=0).
The gate was added when a populated actives array was once seen to empty the
squad. That justification no longer holds:
- it never reproduced, and the feature is now proven end to end on a retail
client - 29 resident nodes, both cardtype-7 kits in club slots 0 and 1 with
itemState 101/102 and category 4, alongside 23/23 players and a resident
manager, with the selector rendering correct distinct home and away kits;
- it can no longer cause durable damage, because both squad write-back paths
are guarded in Core (empty replacement refused; an absent manager field no
longer read as "clear").
Scope audited before flipping: squad_actives() emits ONLY the home and away
kit, each resolved from Core's active designations and required to be genuinely
owned. Badge, ball and stadium are never emitted, so enabling this cannot
surface an unproven active family - confirmed on the wire, where the emitted
itemTypes are exactly {"kit"}.
Env parsing moved into a pure parse_squad_actives() so the DEFAULT is testable
rather than depending on process environment. Unrecognised values stay ON
rather than silently disabling the feature.
Verified on staging with the env var REMOVED from host.env entirely: the host
logs squad_actives=true and serves both kits (assetId 14/15, states 101/102)
with 23/23 players and the manager intact.
club_items.json's _record_map is authoritative: cardassetid is +0x1c and
assetId is +0x20. The probe had them the other way round, which made a correct
assetId 14/15 read out as an identical cardassetid and briefly supported the
wrong conclusion that assetId was not the art selector.
A club item's `assetId` (record +0x20) is family specific and is NOT the
carddbid: per the client's own tables a kit carries the art class from
fcc_kitcards.assetid - 14 for the 63xxxxx home/third band, 15 for the 64xxxxx
away band - a badge carries its team id, and a ball and stadium their own
asset number. The catalog shipped `asset_id`, the carddbid, in that slot.
Measured on the live client with both kits resident: record +0x20 held
6300006 (home) and 6400003 (away) where the table says 14 and 15, while every
other field - resourceId, cardassetid 35, category 2/3, teamid 21, year 0,
itemState 101/102 - already matched. Operator reports both pre-match kit tiles
rendering identically. assetId is the only field that diverges from the
client's own data, and an assetId that is not a valid kit art class cannot
resolve to distinct art.
`resource_id` is derived from `asset_id`, and every home kit shares art class
14, so the two cannot be the same field: catalogs now carry an optional
`club_asset_id`, defaulting to `asset_id` so a catalog predating the field and
every non-club kind are unchanged. resolve_kit emits it as the wire `assetId`.
Fixed at the source too - scripts/sold-staging-up.py emitted asset_id as the
wire assetId for all four club families, so a re-emit would have regressed it.
Field-offset note: club_items.json's _record_map is authoritative and my
earlier working note had these transposed - assetId is +0x20 and cardassetid
is +0x1c, not the reverse.
Adds tools/live/diff_kit_records.py, which byte-diffs the two resident kit
records and names the fields the decoded clone query consumes.
Staging wire now reads assetId 14/15 with cardassetid 35 on both
squad.actives and /club?type=equippables. Workspace 1250 passed, 0 failed.
Client re-parse still to be confirmed visually.
The host called set_squad_manager unconditionally on every squad save, passing
the resolved manager or None. None was serialised as {"owned_card_id": null},
an EXPLICIT removal, so a save that merely said nothing about the manager
deleted the assignment. That is how a client whose squad model had been
destroyed wiped a real manager row (WAL commit 468, squad_managers 1 -> 0).
FIFA 17 has no wire shape that removes a manager: the client always sends a
ref. So None never means "the user cleared the slot" - it means the ref was
absent, zero, or unmappable, i.e. this save carries no manager decision. The
assignment is now left untouched and the skip is logged.
The capability is removed at the TYPE level: CoreAccess::set_squad_manager
takes &str, not Option<&str>, so the host cannot express a clear at all. Core
still supports deliberate removal via an explicit null for other callers.
The existing test asserted the destructive behaviour as intended ("a later save
without a manager CLEARS it"). That contract was the bug; it now asserts the
manager survives and that both saves still commit their slots. With the fix
reverted the test fails.
Live-proven on staging against the real route (PUT /ut/game/fifa17/squad/<n>;
squad/active is a GET-only tail and falls through to the dead Python upstream,
which is why an earlier replay attempt proved nothing):
exact original shape (no resolvable players, manager: [])
-> 502 core_error, core returned status 400, nothing mutated
valid 23-player save carrying manager: []
-> 200 {"id":0}, manager_write_skipped logged, manager PRESERVED
valid 23-player save carrying the real manager ref
-> 200 {"id":0}, manager assigned
Players 23/23, manager 1, captain, active club items, coins, integrity and FK
identical before and after, and again after a Core+host restart.
The squad parser writes actives element i to club-item slot r15d+i, and r15d
is shared scratch that other atom handlers clobber. Which slots are filled
therefore reveals the index the parse actually started from, which is the
open question behind the regression in vault section 18.
probe_resident_fields.py now prints the slot index of every populated entry
plus the empty ones, and verify_kits.sh runs it next to the map census so one
command reports both residency and whether the squad survived.
Populating `squad.actives` is the proven way to make a club item resident —
the squad parser writes each element straight into a club-item slot, both
kits came back resident with the client writing `category 4` itself, and the
pre-match kit selector worked.
But a populated array was then observed to cost the rest of the squad. On a
full client relaunch the resident item map fell from 24 entries to just the
2 kits, the 23-slot player vector came back fully null, and the starting-11
screen was empty; the manager and staff were gone too. Every host response
was 200/outcome=ok with no warning, so the loss is entirely client-side
parse behaviour. `actives` sorts first in the squad object, so `captain`,
`formation`, `manager`, `players` and everything else after it are lost —
consistent with the element loop leaving the tokenizer misaligned.
The same payload produced a correct 24-node map on an earlier relaunch, so
the interaction is not yet understood and is not safely shippable. An empty
squad is far worse than a missing kit, so the array is now gated behind
`HostConfig::squad_actives` (env `OPENFUT_FIFA17_SQUAD_ACTIVES`) and off by
default. The projection, identity plumbing and tests are kept intact: they
are correct and are what the investigation will re-enable.
Staging redeployed with the flag off and verified back to 23/23 populated
player slots and `actives: []`.
The pre-match kit clone resolves BOTH sides out of the client's own
teamkits table keyed on teamtechid, with only teamkittypetechid (0 home,
1 away) telling the strips apart:
teamtechid == record+0x94 (wire teamid)
teamkittypetechid == 0 for activeHomeKit, 1 for activeAwayKit
year == record+0xba (wire year)
Our club wears team 21's kit (fcc_kitcards carddbid 6300006/6400003 are
both teamid 21), and the offline-season ladder cycled a fixed opponent
list whose first entry was also 21. So round 0 put the club against the
team whose kit it wears and both sides rendered the same strip. It is
also simply wrong data: a club playing itself.
The schedule now excludes the club's own kit team, which the host derives
from Core's active kit designations via resolve_kit. An exclusion that
would empty the rotation is ignored, because an empty matches array makes
StartSeason dereference NULL at CardsDLL+0xfc5b5.
This is not a kit-pipeline change: squad.actives already produces the two
resident cardtype-7 records with the correct itemStates, and the clone
query is satisfied by that data unchanged.
FIFA 17 makes a club item resident ONLY through squad.actives. The squad
parser's arm for atom 11 computes the address of the i-th element of the
client's five-element club-item array and hands it to the item
deserializer as the out-handle:
cmp edi,0x5 ; at most five entries are read
mov rax,QWORD PTR [r13+0x108] ; the club-item array
lea rcx,[rax+rcx*8] ; &array[edi]
call 0x18013fe00 ; item deserializer, writing that slot
That deserializer inserts the record into the client's resident item map
- keyed by wire instance id, gated only on the id being non-zero - and
binds the slot handle to it. So each element must be a full item object
like squad.manager[].itemData; an id reference alone installs nothing,
because the manager installer looks its id up in that same map and does
nothing on a miss.
We emitted actives: [] as an 'observed constant', which was circular: it
came from our own captures and the Python oracle seeded it. The client
then read the array-end token immediately, parsed nothing, and left all
five slots null, so every lookup resolved to the static not-found
sentinel whose item pointer is NULL. That is why the pre-match kit
selector had no kits, and it is also why /club?type=kit could never fix
it: no /club response feeds that array.
Core already owns the designations via /club/active-items, so the host
reuses get_active_kits() and the adapter shapes each entry with the same
shape_club_item primitive /club?type=kit uses, keeping one wire dialect.
A Core transport error yields no actives and is reported rather than
silently empty. userInfo.actives already mirrors the squad's.
Verified against the client's own fcc_kitcards table: 6300006 is team
21's home card (category 2) and 6400003 the away card (category 3).
Two tools, both gated on positive controls because the previous pass
produced a confidently wrong negative result.
atom_mapper_emu.py interprets the atom -> field-id dispatch functions
instead of pattern-scanning them. Its selftest encodes the two decoder
traps that caused earlier mistakes - ModRM rm=5 with mod!=0 is [rbp+disp]
rather than RIP-relative, and a constant may reach its use through a
register - plus the live-verified controls that the item mapper maps atom
568 'players' to field id 1 and atom 11 'actives' to 0. The mandatory
manager atom 424 control still fails as a coverage limit: only 2 of the 52
resolver callers are pure dispatch chains, so the tool refuses to support
any absence claim about the squad mapper.
The live probes enumerate the resident item map at owner+0x160c8, whose
layout came from the lower_bound at 0x180119640: key = wire instance id at
node+0x20, record at node+0x28, count at owner+0x160e8. probe_map2 reaches
22 nodes against a count field of 22, so the enumeration validates itself,
and probe_hunt searches all writable memory with its own in-run positive
control. Result: all four club staff are resident, both kit ids are absent
everywhere, and a lookup miss returns the static sentinel 0x1802c2a28
whose +0x10 is NULL - which is exactly the KIT_SCAN symptom.
Records are 0x180 bytes starting at owner+0x162d8, below the embedded
manager subobject at owner+0x1f9d8. Walking that pool answers whether the
client ever builds a record for an item it was served, independently of
whether the record is filed into a collection.
Result on the live client with the kit selector open: 21 records exist -
18 squad players with category 1, plus one cardtype 10 and two cardtype 4
staff with category 0 - and no cardtype 7 record anywhere, despite two kit
items having been served three times in the same boot.
The kit clone driver FUN_1801c3480 gates on record+0x60 == 4, and no
instruction stores that immediate, so the value had to be read off a
genuinely resident record. This dumps cardtype, cardsubtypeid, itemState,
category, teamid and teamkittypetechid for both the player vector and the
club-item vector, resolving the store through the same chain as the census.
Result: all 18 resident players carry category 1 and the five club-item
slots are null, which identifies +0x60 as a per-collection tag rather than
item data.
A client-local nat rule redirecting dport 8081 to the plain-HTTP staging
UTAS host captures the roster/squad-update TLS connection, which is
https://winter15.gosredirector.ea.com:8081/fifa17/fut/rosterupdate.xml.
The staging host completes the TCP handshake then closes on the
ClientHello, so the client retries and shows "An error occurred
downloading the FUT squad update."
Proven by capturing on the server while probing from the client: a
connection addressed to :8081 arrives as dport 8299, while an :8443
control in the same capture yields 75 packets and a clean handshake.
The script deletes only nat rules whose destination port is 8299, then
verifies the endpoint presents CN = winter15.gosredirector.ea.com.
Read-only watcher that waits for FIFA17.exe, re-resolves the club-item store
each tick (the manager is reallocated per login), and emits one timestamped line
per CHANGE. Pair it with
journalctl -u openfut-staging-host -o short-iso
to answer "after which response does a resident club item first appear?" by wall
clock, without having to reverse the constructor first.
Re-resolving per tick matters: the store is reached through
[CardsDLL+0x2e6398] -> vtable[0x4e8], and that getter is a `lea`, so the manager
is an embedded subobject whose address moves with the owner. The tool also
re-checks the module base against a known immediate every time it attaches and
refuses to report from a wrong base.
Verified against the currently live client: club=0/5 players=18/23.
Staging can host the session: every bootstrap route probed answers 200
(userMassInfo, squad/active, squad/list, club/stats/*, hub, user, club arms,
item idList, clubUser, season/list, watchList, purchased/items, settings,
clientdata) with the single exception of /statistics/tournament, which 502s
because staging's Python upstream is deliberately dead and which is not on the
club bootstrap path.
Read-only /proc/PID/mem census of the client's resident club-item store, which
is what the pre-match kit selector actually consumes. No writes.
Resolves the whole chain from static RE rather than guessing offsets:
[CardsDLL+0x2e6398] -> owner object (FUN_18011a830 is a plain
global read)
owner->vtable[0x4e8] -> lea rax,[rcx+0x1f9d8]; ret, i.e. mgr is an
EMBEDDED subobject, not a pointer
mgr+0x108 .. mgr+0x110 -> club-item vector, stride 24
element+0x10 -> the item record (FUN_1800d73d0)
CardsDLL is located by its NEAREST PRECEDING NAMED mapping, because Wine maps PE
sections anonymously and the Wine heap is also rwx, so permissions do not
discriminate code from heap. The base is then sanity-checked against a known
immediate (mov edx,0x7575 at 0x180026fea) and the tool aborts rather than
reporting from a wrong base.
Field offsets are the ones already proven, and nothing else is interpreted:
+0x4c cardtype, +0x50 cardsubtypeid, +0x5c itemState, +0x60 category,
+0x94 teamid, +0xba teamkittypetechid (u16).
FIRST RESULT, live on the client parked at the pre-match kit selector:
players mgr+0x0d8: 23 slots, 18 non-null, all (cardtype 1, subtype 0)
club items mgr+0x108: 5 slots, 0 non-null
Five slots, every item pointer NULL. Five is the club's active club-item set --
home kit, away kit, badge, stadium, ball -- so the client knows it should hold
five and holds none. That is why FUN_1800d73d0 returns the 0x1802c2a28 sentinel
whose +0x10 is NULL, and why the tiles are untextured.
Recovered from the live retail client (pid 44405, parked on the pre-match kit
selector) by read-only /proc/PID/mem. No writes, no debugger, no detours, no
client modification. Denuvo decrypts FIFA17.exe in memory, so the immediate that
is absent on disk is present at runtime.
At live 0x147de80e8 FIFA17.exe runs a name-registration loop through the same
string-setter vtable slot (+0x20) CardsDLL uses, so each id can be given its
name:
0x7579 useSavedMatchData -> movb $1, ctx+0x151
0x757a useSavedMatchKits -> movb $1, ctx+0x152 == KITS_AVAILABLE
0x7587 setFUTServerEnvironment
271 pairs recovered and committed. This is a DIFFERENT namespace from CardsDLL's
DataProvider id table -- the same numeric id has a different name in each --
matching the APT's own split between game.uif.UIFDataProviderList and the
action/command list.
KITS_AVAILABLE therefore does not mean "the server sent kits". It means "use the
previously SAVED match kits", and the APT writes those via ACTION_SAVE_MATCH_KIT
in SaveKitsForMatch. Live, both flags read 0 on a fresh match at the exact
moment the tiles are blank. So 0 is CORRECT on first entry and the client is
designed to take the fallback path. The gate was never the defect, which is what
the 2026-08-23 measured negative was already indicating when better item data
changed nothing.
The fallback is GetKitArrayForFUT -> ION_Uniform.GetIDs(teamId), which natively
gates on team id 130000 (0x1800d8ab0) or 130001 (0x1800d8ad0) -- the synthetic
FUT home/away pair -- and otherwise defers to the generic engine catalogue. For
a matching team it packs the two active kit records and admits them through a
check that resolves at runtime to "cmp edx,0x189a1003 / setne al", i.e. merely
"not the invalid sentinel".
Still no OpenFUT change: whether the selector requests 130000/130001, and
whether the active kit records resolve, are both unanswered and both live
questions. Full write-up in the Vault under Kit Selector APT Decode section 8.
d146f9c flipped `club?type=equippables` from withheld to answering the KIT
family — the change that made kits render in My Club, operator-confirmed — but
it only touched src/lib.rs. `withheld_club_type_arms_are_empty_with_a_recorded_reason`
still listed that arm, so the suite asserted the exact behaviour the fix
removed. It failed on `cargo test --workspace`; the "1241 passed" figure carried
in the notes was measured before the flip and was stale.
Drops the arm from the withheld table and adds a test for what actually ships:
the tab answers, serves exactly the kit family with cardsubtypeid 9, and returns
the same body as `?type=kit`, so the two tabs cannot drift apart.
`OPENFUT_FIFA17_EQUIPPABLES=0` is deliberately not exercised: it is read through
a OnceLock, so flipping the process-wide env in one test would leak into every
other test in the binary.
Adds fifa17-recon/tools/apt_decode.py, a clean-room decoder for the EA APT
compiled-ActionScript format as FIFA 17 ships it, and deletes avm1_disasm.py,
which assumed SWF framing and could not decode this artifact.
FORMAT. FIFA 17 uses a 64-bit variant of the format: constant-pool entries are
16-byte {u64 type, u64 value} in a separate "Apt1" container member, container
pointers and counts are u64, DefineFunction2's operand block is 48 bytes rather
than 28, its 0x1234567898765432 trailer is stored as two u64 halves, and
parameterised instructions align their operand block to EIGHT bytes, not four.
The alignment is the fact that made the stream decodable: the ConstantPool at
0xd38 yields garbage at align-4 and count=401 at align-8, with an index array
that terminates exactly on the parameter-list region.
The three opcodes that blocked the previous attempt are all unaligned 2-byte
records: 0xAF EA_GetNamedMember (u8 constant-pool index, pop object push member),
0xB9 EA_PushRegister (u8 register index), 0xA2 EA_PushConstantByte (u8
constant-pool index). Byte-wide indices only reach pool entries 0-255, which is
why CheckIsKitLocked at index 293 is emitted as 0xA3 PushConstantWord.
Format facts came from a written specification derived from OpenSAGE
(588ac477367a0022adf29f20a084e8873014e6ce, GPL-3.0 with EA section 7 additional
terms). No code was copied or transliterated; only the interface specification
was used. Provenance is recorded in the module docstring.
VALIDATION. The whole artifact decodes: 5251 instructions, 24 functions, action
stream 0xd38..0x3e75 with 12605 of 12605 bytes covered and zero interior gaps,
zero unresolved opcodes, zero invalid branch targets, zero unresolved strings.
Every byte of the 20630-byte member is accounted for by region. --selftest
asserts all of it, plus operand fixtures and fail-closed behaviour on truncated
records, out-of-range pool indices and unknown opcodes. Unknown opcodes still
refuse to guess a length rather than resynchronising.
RESULT. KITS_AVAILABLE is a BOOLEAN in the DataProvider header, not a count, so
the logged Some(0) means false. futSelectTeam::Publish reads
publishObject.header.KITS_AVAILABLE == true and only then fills m_arrKitPanelData
from data[side].LENGTH and data[side]["KIT_"+i]; CardsDLL's builder writes
exactly that shape ("LENGTH" and "KIT_%d" confirmed in .rdata). The flag comes
from ctx+0x152, whose only setter is internal message 0x757a. That message is
never constructed in ActionScript (the asset contains zero integer literals
above 255) and never sent by CardsDLL (247 of 247 send sites pass an immediate,
none of them 0x757a; four sites in the same family are the positive control).
So no HTTP response can open this gate, and no OpenFUT change is made here.
CheckIsKitLocked is called at 0x2cc3 but defined on the mcSelectTeam child clip
in another asset, so its body is deliberately NOT reconstructed rather than
guessed. Full findings in the Vault under Kit Selector APT Decode.
Operator exported futSelectTeam.BIG (88272 B, BIGF, 11 members). Confirmed the
right screen: FUT_GET_MATCH_KITS_DP, CheckIsKitLocked, mcLockHome and
KITS_AVAILABLE all present. Members split out and committed:
futSelectTeam_Apt1.bin 14526 B magic Apt1
futSelectTeam_AptData.bin 20630 B magic "Apt Data:1:7:8"
STRUCTURE, measured rather than assumed:
* Apt1 is header + a u32 STRING POINTER TABLE + an 8-byte-aligned string table.
Every symbol has exactly one u32 reference and the refs run in the same order
as the strings, so they are pointers, not code references.
* Apt Data holds the actions. Opcode frequencies are AVM1-shaped - GetMember
0x4e x309, If 0x9d x172, CallMethod 0x52 x160, DefineFunction2 0x8e x33,
Jump 0x99 x74 - but two non-standard opcodes dominate (0xaf x1081,
0xb9 x1064), so this is EA's extended dialect and strings are referenced by
table offset instead of a SWF ActionConstantPool.
Adds avm1_disasm.py, which reads the standard subset and prints unknown opcodes
with their length rather than skipping them. It is NOT sufficient for this file:
decoding 0xaf/0xb9 is the remaining work, and OpenSAGE apt-toolkit is the
reference implementation for EA APT actions.
So CheckIsKitLocked is located but not yet READ. What is known stays known: the
native side only ever writes LOCKED = 0, so the predicate lives here.
The Frostbite .cas chunks holding APT ActionScript are greppable, so screens can
be identified and their whole symbol table recovered without driving the GUI.
Control: KitAssignmentPopup (a string from an already-exported BIG) hits 43 times
across the 52 cas files, so a miss would have been meaningful.
FUT_GET_MATCH_KITS_DP hits 10 times. The binding screen is
external.ion_fut.screens.futSelectTeam
(fifa_installpackage_01/cas_01.cas @ 0x3707ecd7), which no exported BIG contained
after 33 attempts at guessing names.
It binds FUT_GET_MATCH_KITS_DP, KitSelectDP and TeamSetupDP, and carries exactly
the vocabulary the native side implies:
panels/locks mcKitHome mcKitAway mcLockHome mcLockAway m_arrKitPanels
sides HOME_SIDE AWAY_SIDE NEUTRAL_SIDE SIDE_HOME SIDE_AWAY
DP fields KITS_AVAILABLE KIT_ HOME_KIT_ID AWAY_KIT_ID
flow InitializeKitConfig GetKitArrayForFUT InitializeKitsFromArray
EnterKitSelect IsKitSelectCreated ExitKitSelect SaveKitsForMatch
lock CheckIsKitLocked RemoveKitLocks SetKitReady SetKitUnReady
uniform SetUniform ION_Uniform GetNonConflictingUniformID
CheckIsKitLocked is the lock predicate the native side does not own — recall
sub_180033430 only ever writes LOCKED = 0, so the "kit is currently locked" dialog
is raised here.
Adds find_apt_in_cas.py (control-guarded) and the recovered 934-symbol table.
OPERATOR-CONFIRMED on the retail client: the My Club kit tab shows both kits, the
first time kits have ever rendered in this project.
The tab asks for `?type=equippables`, and that arm was withheld, so the screen got
an empty body and showed nothing. That is exactly what "0 kits in my club" was —
not a shaping problem, a refused question. Log line that identified it:
route=club outcome=withheld filter=[type=equippables,
reason=multi_family_crash_2026_08_05] total=0 emitted=0
TWO fixes were both required, so neither alone is the cause:
* this arm answers (KITS ONLY), and
* GET ut/%s/item (FutViewCards) returns the OWNED INSTANCE rather than a
definition placeholder, so the kit arrives as cardsubtypeid 9 /
itemState active{Home,Away}Kit instead of "a free player" (d4a39ba).
The 2026-08-05 crash that motivated the withholding was 30 items across FIVE
families. This arm serves one family, and the live body is two items. Serving the
other four families here is untested and stays unserved.
Default flips from opt-in to opt-out: OPENFUT_FIFA17_EQUIPPABLES=0 restores the
withheld body with a restart and no rebuild. The switch is kept rather than
deleted because the only live evidence is a club holding exactly TWO kits; a club
holding many is untested and the original crash was about size and family mixing.
Verified with no env flag set: ?type=equippables -> 2 items, family {9}.
cargo test -p openfut-utas-host: 190 passed, 0 failed. clippy -D warnings clean.
Production untouched — staging only.
GET ut/%s/item is FutViewCards and its ids are OWNED INSTANCE ids - the client
builds the query as ?idList=%lld (CardsDLL .rdata 0x220080) from ids it already
holds. It was being answered with the definition body, which echoes the queried
id straight back. Asked about the active home kit, instance 100004874, the server
replied:
resourceId 100004874, cardsubtypeid 0, itemType "player", itemState "free"
i.e. "your active home kit is a free player". No kit can ever be seen as active
through that. Real players were equally wrong: instance 100000003 came back as
resourceId 100000003 instead of the actual card 200389 rating 87.
This is on the active-kit path, and the evidence for that is the client's own UI.
KitAssignmentPopup.BIG (exported from Frosty today) decompiles to
external.ion_fut.components.KitAssignmentPopup and contains OSDKCards_ViewCards,
OSDKCards_ActivateCard, mHomeKitID/mAwayKitID/mSourceKitID, and the search states
SEARCH_STATE_ACTIVE_HOME_KIT / SEARCH_STATE_ACTIVE_AWAY_KIT. Those states can
only come from the itemState this route returns.
Route::ViewCards is now distinct from Route::ItemDefs. Owned instances are shaped
by the SAME projector /club uses, so there is one wire dialect and no drift; ids
that are not owned instances still fall back to the definition placeholder, and
the empty query still answers {"itemData": []}, preserving oracle parity for the
definition-style callers (item/resource, defid).
Verified on staging:
?idList=100004874,100004873 -> resourceId 6300006/6400003, cardsubtypeid 9,
itemType kit, itemState activeHomeKit/activeAwayKit, teamid 21, cat 2/3
?idList=100000003 -> resourceId 200389, rating 87 (the real card)
?idList=999999999 -> definition fallback
no query -> {"itemData": []}
item/resource? and defid? -> unchanged
cargo test -p openfut-utas-host: 190 passed, 0 failed. clippy -D warnings clean.
Fifth instance of this project's recurring dead-route defect: a handler exists and
is correct, but classify() never produces the route, so every request falls
through to the Python upstream. Invisible in production, where the oracle answers;
on staging, where the upstream is deliberately dead, it is a 502.
GET ut/%s/item is FutViewCards (deser 0x1801293d0, top-level itemData via the
shared card element 0x18013fe00). The oracle answers it with defs_route - the SAME
handler already wired for item/resource and defid: pull every integer out of the
query (idList=a,b,c / definitionId= / resourceId=) and return one definition per
id, or {"itemData": []} when there are none (tools/utas_server.py:1103). So
claiming it is byte-identical parity, not new behaviour.
The query handling is the substance of the fix, not decoration. ut_tail does NOT
strip the query string, so an equality-only arm (`Some("item")`) misses every real
request while passing a no-query unit test - which is precisely how the route
stayed unclaimed. The same latent bug applied to the two arms that were already
there: item/resource and defid only matched with no query, so
`item/resource?resourceId=` and `defid?definitionId=` were ALSO falling through.
All three now mirror the oracle's own `item(\?|$)` pattern.
Verified on staging, all 200 where they were 502:
GET /item -> itemData 0
GET /item?idList=100000003,100000004 -> itemData 2
GET /ut/v2/game/fifa17/item -> itemData 0
GET /item/resource?resourceId=5003012 -> itemData 1
GET /defid?definitionId=200389 -> itemData 1
Does NOT by itself fix the kit selector - GET /item is a definition lookup keyed
by ids the client already holds, not the thing that seeds the club collection, and
the observed session never requested it. It is a real production-masked gap and a
prerequisite for testing anything else on staging.
get_item_is_claimed_and_the_other_item_verbs_are_unaffected pins the claim plus
the three verbs that share the prefix: PUT item stays FutMoveCard on the economy
path and DELETE item/<id> stays QuickSellPath.
cargo test -p openfut-utas-host: 190 passed, 0 failed. clippy -D warnings clean.
RETRACTION FIRST. The previous commit added itemType to club items on the theory
that it gated ingestion, because player/staff sent it and were resident while
kit/badge/stadium omitted it and were not. Relaunched the client with itemType on
all three: ?type=kit answered total=2 emitted=2, and still no cardtype-7 record.
The correlation was an artefact of the control. Measured read-only over
/proc/PID/mem with full coverage (3605 MiB, nothing skipped): the "resident"
players and staff were all SQUAD members, which arrive via userMassInfo. Players
that appear in /club?type=player but NOT in userMassInfo are not resident either -
0 records for 6 of 6 sampled, 5 with no byte match at all, out of 1966 served.
Residency tracks the ROUTE. /club?type= responses never enter the persistent card
collection, and no value of itemType changes that. itemType is kept as wire
fidelity (every real EA item carries it) and relabelled; its doc no longer claims
to fix anything. The diagnostic KIT_PROBE is removed - it could only have tested
shape hypotheses that this result makes moot.
RECOVERED from the unpacked CardsDLL, no archive extraction, no instrumentation:
Packed kit id, both directions present and agreeing:
id = (teamid << 14) | (year ? (year-1800) << 5 : 0) | kittype
so a kit is addressed by the triple (teamid, year, kittype).
FUN_180033770 answers ONLY for team 130000 - 0x1800d8ab0 is literally
`mov $0x1fbd0,%eax ; ret`. Every other team id falls through to the engine's
catalogue kits, which are the lockable ones.
sub_180033430 writes the tile: NAME = "HOME_SIDE"/"AWAY_SIDE", TYPE = the
localised Kit_type_0 / Kit_type_1 / Kit_type_historical, and LOCKED (always
value 0, never 1). If the queried triple matches NEITHER active triple it
writes NOTHING - which is exactly why one tile rendered "undefined". A missing
write, not a bad string. There is no Kit_type_2.
FUN_1800d73d0 selector 2/3 does `setne dil ; add $0x65,%edi` then compares
itemState: active home = 101, active away = 102, derived arithmetically and
independent of the enum table. year at +0xba is movzbl - a byte INDEX.
Above all of it: FUT_GET_MATCH_KITS_DP (0x7565) handler FUN_1800be6a0 gates on
`cmpb $0x1,0x152(%r14)` and returns early otherwise. KITS_AVAILABLE IS
ctx+0x152. Constructor zeroes it; the only setter is case index 6 (message
0x757a) of the jump table at 0x1800c00d4. Live value is 0, so no kit list is
ever built. 0x757a has no name in CardsDLL and that is bounded, not sloppy: the
registration run ends at 0x7575 with the epilogue immediately after, and 70
other ids resolve from the same table as the positive control.
Tables (audit_fifa17_kits.py, full-table counts): category 2/3/5 -> engine kit
type 0/1/2 with 0 counterexamples against 54/166/145 discriminating keys; the id
band is NOT home/away (band 63 holds 740 home AND 88 third).
Vault: "Kit Selector Data Path.md". cargo test 429 passed 0 failed across the two
crates; clippy -D warnings clean; fmt clean.
Answers from the extracted client tables, before touching a binary. Every number
is a count over the full table.
fcc_kitcards 1482 rows, teamkits 2576 rows.
CATEGORY -> ENGINE KIT TYPE, with a test that can actually fail. Asserting
"category 3 means away" because away kits usually exist is not evidence: types
0/1/2 are present for most teams, so it is true by construction. The
discriminating cases are the teams that LACK a type.
category 2 -> type 0 HOME 54 keys lack type 0, 0 counterexamples
category 3 -> type 1 AWAY 166 keys lack type 1, 0 counterexamples
category 5 -> type 2 THIRD 145 keys lack type 2, 0 counterexamples
and there is never more than one card per (team, year, category).
THE ID BAND IS NOT HOME/AWAY. Band 6300000 holds 740 HOME cards AND 88 THIRD
cards; band 6400000 holds the 654 AWAY cards. assetid is fully determined by the
band (14 for 828/828 of 63xxxxx, 15 for 654/654 of 64xxxxx), so it carries no
information the band does not. cardassetid is 35 on all 1482 rows - it is the FUT
card frame, not the kit art.
This matters for openfut-adapter-fifa17: KIT_AWAY_FLOOR splits home from away at
6_400_000, which is right for home vs away but silently classifies all 88 THIRD
kits as HOME. Recorded here, not yet fixed - third kits are not currently
ownable, so nothing observable depends on it.
teamkits.islocked is 0 on all 2576 rows, so the DB lock flag is NOT what makes
the pre-match selector call a kit locked. 6 rows are embargoed.
Team 21, the staging club, resolves exactly:
6300006 cat 2 HOME year 0 assetid 14 -> teamkitid 1376
6400003 cat 3 AWAY year 0 assetid 15 -> teamkitid 1377
6300007 cat 2 HOME year 1972 assetid 14 -> teamkitid 5126
6300008 cat 5 THIRD year 0 assetid 14 -> teamkitid 1378
so the two kits OpenFUT serves are the correct home/away pair.
NOT recoverable from data/tables: the kit's own name string. fcc_kitcards
name/header/description/biodescription are byte OFFSETS into the table's string
blob (583, 597, 608, 619 on one row), and that blob is not among the extracted
tables.
Adds audit_fifa17_kits.py (the tool, with the discriminating test inline) and
fifa17-kit-map.json (its output) so this is reusable data rather than terminal
scrollback.
The submodule pointer still referenced the rate-limited engine-provider build.
d764117 removes those detours entirely; they froze the client once and crashed
it once, and the fault was the detours themselves rather than their logging.
RESULT OF THE PREVIOUS COMMIT, recorded before anything else: sending itemType on
cardtype-7 club items did NOT make them ingest. Client relaunched, ?type=kit
answered total=2 emitted=2 with itemType="kit" on both, and afterwards there is
still no cardtype-7 record resident and the hook still traces
KITS_AVAILABLE = 0. The player/staff-vs-kit/badge/stadium correlation was real
but it is NOT the cause. The field is kept because every real EA item in the
capture corpus carries it and the two ingesting families already did, but it is
now labelled wire fidelity, not a fix.
Also already refuted, so neither is the answer: ?type=equippables answered with a
kits-only body (kits stayed undefined), and the kit ids are correct - 6300006 and
6400003 are the real fcc_kitcards carddbids for team 21 home/away with matching
category 2/3 and year 0.
This adds OPENFUT_FIFA17_KIT_PROBE (default OFF, staging armed) which appends two
synthetic kits to ?type=kit so ONE client restart discriminates the two remaining
hypotheses instead of one restart each:
6300007 MINIMAL - exactly the field set a STAFF item carries, which is known to
ingest, plus cardsubtypeid 9. If only this appears, one of the kit-only
extras (assetId, cardassetid, teamid, category, year) makes the client
discard the item.
6300008 NAMED - full kit shape plus name/localizedName/description, the three
fields the cardtype-7 parse arm is documented to copy and which OpenFUT has
never sent. If only this appears, they are required, not optional.
If NEITHER appears, ?type=kit is not the route that populates the collection
FUN_1800d73d0 scans, and the search moves to which route does.
Both ids are real team-21 carddbids, served free so they cannot disturb the
active-kit assignment, with instance ids outside Core's range. Two items in one
family: the response that crashed this client on 2026-08-05 was thirty across
five.
Measured live 2026-08-23 against the client parked on the pre-match kit selector
(pid 8793, read-only /proc/PID/mem). Whether a served club item becomes a
resident item record correlates perfectly with whether we send itemType:
family itemType sent resident record?
player "player" yes
staff "staff" yes
kit (absent) NO
badge (absent) NO
stadium (absent) NO
Two of two families that carry it are ingested; none of the three that omit it
is. With no cardtype-7 record resident the club scan FUN_1800d73d0 matches
nothing, the FUT match-kit DataProvider is built empty (traced: KITS_AVAILABLE
= 0), and the selector falls back to catalogue-gated engine kits - which is the
"This kit is currently locked. To unlock and use it, please go to the Football
Club Catalogue." dialog the operator sees. That string lives in the Flash UI, and
Blaze receives no catalogue or unlock request at any point, so the lock is
decided client-side from data the client already holds.
CARD_SYSTEM.md records that itemType "is parsed into a heap string and never
stored". That stays true of the RECORD; it does not follow that the string is
unused, and the correlation is evidence it is consulted before the record is
kept.
Tokens come from the ?type= vocabulary decoded from the FUN_18012ec50 jump table
(kit 12, stadium 13, badge 11) - the same vocabulary the two working families
use. INFERRED, not proven: no real EA club item exists anywhere in the capture
corpus, so the exact token is taken from the atom vocabulary, not observed wire.
Near miss worth recording: STADIUM_SUBTYPE was not imported at module scope, so
match treated it as a fresh binding instead of a constant, made that arm
irrefutable and typed badges as "stadium". It compiled with only an
unused-variable warning. every_cardtype7_family_carries_its_own_item_type
asserts three distinct tokens, which is what catches that.
cargo test --workspace: 1240 passed, 0 failed.
The 2026-08-21 entry concluded the pre-match kit selector "is a client dead end,
not a missing wire field" because nothing stores 4 into item +0x60. Withdrawn.
It rested on two mistakes:
1. +0x60 == 4 DOES occur live - a record with +0x4c == 2 and +0x60 == 4 reached
the art-clone driver FUN_1801c3480. The immediate-store scan cannot see it,
so "nothing can satisfy the gate" was never licensed by that evidence.
2. It annotated `cmp [rdi+0x4c], 7` with "<- we produce this" without measuring
it. Its own live half showed only {1: players, 0: staff}: zero cardtype-7
records. That is the finding, and it was read as the opposite.
Measured now against the live client parked on the kit selector, read-only via
/proc/PID/mem over 3047 MiB, searching the exact u32 values the server sent:
players and staff are resident with sane fields; kit, badge and stadium are all
absent by resourceId AND by instance id. The host served ?type=kit total=2
emitted=2 at 17:50:09 this session and neither kit produced a record.
So the blocker sits upstream of the +0x60 gate: no cardtype-7 record is ever
created, so the club scan FUN_1800d73d0 has nothing to match, KIT_DESC never
fires and KITS_AVAILABLE reads 0. Cause is not yet settled - either our wire
shape (the cardtype-7 arm wants name/localizedName/description, which we do not
send) or cardtype-7 items being transient. Neither is recorded as fact.
Also: kit_gate_probe.py's live half is unreliable. On pid 8793 it reported
"CardsDb is empty" while a byte scan found 1966 resident players, so its
structural chain is stale and its record counts understate reality. Adds
club_record_residency_probe.py, which is read-only and cannot disturb the game.
A live kit_trace run on the retail client showed the kit clone driver only ever
sees PLAYERS - about 19 records, all cardtype 1 / itemState 1 / +0x60 = 1 - and
never a kit, with no KIT_DBCLONE line at all. So the locked-kit failure is
UPSTREAM of the item+0x60 == 4 gate that this arm's comment blamed.
In the same session the client asked for ?type=kit six times, which we answer
and which feeds the items browser, and ?type=equippables twice, which we answer
empty. If the equippable view is what populates the collection FUN_1800d73d0
scans to build the active-kit triple, an empty answer explains precisely why the
triple stays zero and the engine falls back to its own catalogue kit.
The arm now selects ContentKind::Kit behind OPENFUT_FIFA17_EQUIPPABLES=1, so it
answers with TWO items rather than the thirty across five families that crashed
the client on 2026-08-05. Default OFF: that crash is real and reproducible, and
the narrow body is a hypothesis under test rather than an established safe
response. Reverting is an env change plus a restart, no rebuild.
Operator report: selecting a kit in the pre-match selector gives "This kit is
currently locked. To unlock and use it, please go to the football club
catalogue."
Root cause, from static RE of the UNPACKED CardsDLL. CardsDLL registers a kit
provider into the FIFA engine (singleton FUN_1800338f0, vtable 0x1801f1d68):
slot +0x08 FUN_180033770 enumerate kits for a team
slot +0x10 sub_180033430 describe one kit <- the lock gate
The describe function decodes a packed kit id into (teamid, year, slot) and
compares it against the club's ACTIVE HOME and ACTIVE AWAY triples. On a match
it writes NAME/TYPE (and, for historical kits, LOCKED). On no match it writes
NOTHING AT ALL and the descriptor falls through to the engine's own default,
which is where the locked-catalogue message comes from.
The active triple is 100% server-driven. FUN_1801c26d0 -> FUN_1800d73d0 scans
club items for cardtype 7 + cardsubtypeid 9 + itemState 101/102, then reads:
teamid <- item+0x94 (atom 0x306) we were sending this
year <- item+0xba (atom 0x389) WE WERE NOT SENDING THIS
slot <- item+0xb8 (category) WE WERE NOT SENDING THIS
category 2 -> slot 0 home, 3 -> slot 1 away, 5 -> slot 3 third
So we shipped kits carrying only teamid, and the triple could never match.
fifa17-recon/data/club_items.json already has category and year per kit
resourceId (1482 kits; category {2:740, 3:654, 5:88}; 85 historical years), so
this is carried through the catalog rather than invented: Fifa17CardIdentity
gains category/year, Fifa17KitIdentity carries them, and shape_club_item emits
them for KIT_SUBTYPE only. Badges and stadiums deliberately do not gain the
fields - the slot mapping is kit-specific, and sending a family a field its
resolver does not read is how this project previously froze the client.
Also corrects the Vault note: item+0xba is `year`, not `kittype`. The side comes
from itemState 101/102 and the slot from category.
Two things this does NOT fix, both client-owned and recorded rather than
guessed:
- the runtime teamkits clone into FUT club 130000 (the kit ART) is gated on
item+0x60 == 4, and +0x60 has no wire atom at all: the deserialiser
unconditionally zeroes it, the only CMP against 4 in the whole DLL is
0x1801c34f1, and a live probe measured it as 1 for players / 0 for staff,
never 4. The existing lib.rs claim that a server can never produce 4 is
CONFIRMED, though its stated reason was a live observation rather than the
real one (no wire atom exists).
- whether a year==0 kit needs an explicit LOCKED write is UNKNOWN: CardsDLL
only writes LOCKED for year != 0, and the engine's default for an untouched
descriptor is in Denuvo-packed FIFA17.exe.
The client fetches the roster from the URL our own Blaze hands it,
https://winter15.gosredirector.ea.com:8081/fifa17/fut/rosterupdate.xml, and
ProtoSSL verifies that certificate by dNSName only. An IP-addressed roster host
is refused even with IP Address:10.10.0.120 in the SANs (retested on Windows
2026-08-23), so the hostname has to survive into SNI while the connection lands
on us.
The hook log showed the dial was ALREADY being intercepted:
connect_hook: call 20.51.153.159:8081 (octets logged reversed)
It simply was not rewritten, because 8081 was not in the EA port table. So this
is a table entry, not new hook surface: ea_ports::FIFA17_ROSTER makes the
existing connect/WSAConnect/ConnectEx detour rewrite the destination while
leaving the URL untouched, and the certificate still validates.
That removes the scoped DNS responder and the client NRPT rule from the normal
path. scoped-dns.py stays as the documented contingency for EA withdrawing the
public record, which is the one dependency this cannot remove: the name must
still resolve to something for connect() to be reached at all.
roster_port is accepted in openfut.cfg but written only when non-default. The
parser rejects unknown keys, so emitting it unconditionally would make every
already-deployed hook reject the file and install NO redirect, breaking the game
rather than degrading.
Also corrects two documented claims that are false on the real machine:
elevation comes from the shortcuts' RunAsAdmin bit, not from an AppCompatFlags
RUNASADMIN entry (there is none), and hook_dll_path must point at a source copy
rather than the deployed version.dll it is copied onto.
FIFA 17's ProtoSSL verifies the roster certificate by dNSName only, so the
client must reach the roster as winter15.gosredirector.ea.com. Retested on
Windows 2026-08-23: an IP-addressed roster host is refused even though the
certificate carries IP Address:10.10.0.120 as a SAN.
Public DNS points that name at EA's dead 159.153.51.20, so the client has to
resolve it to us. scoped-dns.py pins exactly that one name and forwards every
other query upstream verbatim, so a client pointed at it cannot lose general
resolution -- verified against www.microsoft.com, github.com and
www.msftconnecttest.com.
Paired with a Windows NRPT rule rather than a hosts entry: per-name, auditable
via Get-DnsClientNrptRule, and revertible in one command. A hosts edit on this
machine had previously taken its whole internet down.
Passes a null attribute slot for the rare card -- Core reads absence as
"every slot", so sending 0 would silently train pace alone -- and declares
the per-family ceiling rather than a single constant.
Tests pin the null-slot serialisation, both ceilings, and that all 36
single-attribute plus all 6 rare cards resolve.
They were failing closed as "squad fitness". Four facts say otherwise: each
family holds exactly 21 rows = 7 types x 3 levels matching the published
"6 single attributes + 1 ALL" list; those six rows are the ONLY ones with
weightrare=2 while all 36 single-attribute rows are 0, and the published list
marks ALL rare; their amounts are exactly 3/6/10 against the documented ALL
card's +3/+6/+10; and bc=6 is one past the six real slots, an "all" sentinel,
with c0=0 reading as "not single-ATTRIBUTE" rather than "not single-target".
The misreading came from consumables.json's FUT_FITNESS_UC/MC label, which is
tool-authored -- build_consumables.py names the 7th element of its attribute
array -- and has no documented provenance. The bc/c0 values beside it ARE
reversed; only the name was not. The genuine squad-fitness card is subtype
220 in fcc_healingcards (10/20/30) and player fitness is 219; that family
stays unsupported.
The two families carry different authored ceilings, 15 single-attribute and
10 all-six, so ceiling_for() reports the right one per effect -- sending the
single-attribute ceiling for an all-six card would let a +15 all-six boost
through, granting 90 attribute points from a card worth 60.
Also corrects ENDPOINT_MAP row 7: ApplyCardByRes carries urlIndex 0x0e, which
resolves to ut/%s/item/resource, and the live verb is POST. The row claimed
PUT ut/%s/item for both apply RPCs, and that conflation is what kept the
"apply must ride PUT ut/%s/item" hypothesis alive until the POST capture
settled it -- every observed PUT ut/%s/item is a pile move.
Opens the 409 `apply_effect_unproven` gate for attribute training, the
second family after contracts to have its effect settled rather than merely
its magnitude.
`ApplyEffect` replaces the single-family `AddContractMatches` struct: Core
dispatches on `kind`, so an unproven family must be impossible to express,
not merely discouraged. The shared half of an apply -- the exactly-once key,
Core's transaction, the error mapping and the client's payload -- is now one
`finish_consumable_apply`, so a new family cannot quietly acquire its own
idempotency format or its own success shape.
Two refusals are training-specific and both prevent silent corruption rather
than merely being tidy: a non-player target has no attributes to write, and
a cross-class target would train a different attribute from the one printed
on the card, because a keeper's slots mean DIV/HAN/KIC/REF/SPD/POS where an
outfielder's mean PAC/SHO/PAS/DRI/DEF/PHY.
`attributeList` now prefers Core's `effective_attributes` and only falls
back to the immutable definition when Core does not send them -- reading the
definition regardless would silently drop every applied training off the
card the client draws.
Tests pin the verb split on `item/resource/<rid>` (POST applies, PUT stays
quick-sell, GET is not an economy route at all), the digit guard, the exact
JSON Core deserialises for both effects, and that no shipped training card
exceeds the ceiling the host declares to Core.
Which attribute a FIFA 17 training card trains is recoverable after all --
not from a table, but from the client's own dispatch: `FUN_18013f4d0`
derives a consumable's whole presentation from `cardsubtypeid` and writes an
attribute selector to `rec+0xbc`. Paired with `fcc_trainingcards.amount`
(TABLE_PROVEN, matching the wire 8/8), that settles both halves of the
effect for all 36 attribute training cards -- 12 families x 5/10/15, 18
goalkeeper and 18 outfield.
The subtype order is NOT the slot order: 54 is SPEED at slot 4 while 56 is
REFLEXES at slot 3, and 65 is HEADING at slot 5 while 66 is DEFENDING at
slot 4. Reading them sequentially trains a different stat from the one on
the card, invisibly, so the table is explicit and a test pins those four.
The two SQUAD training cards (57, 67) share the table and the client's
training UI bucket but are the only ones whose single-target byte is 0: they
act on a squad and move fitness, not an attribute. They resolve to None and
fail closed rather than being mistaken for a +3 attribute card.
A keeper's six slots mean DIV/HAN/KIC/REF/SPD/POS and an outfielder's mean
PAC/SHO/PAS/DRI/DEF/PHY -- same numbers, different attributes -- so the
class gate is not cosmetic.
The reboot-survival gate is a test no operator can stand inside: the machine
under test is the machine running the session. So the machine records its own
recovery.
openfut-boot-evidence.service polls until the anchor, Core and host agree on
a namespace (or a 180s deadline expires), then writes a JSON file with the
boot id, anchor/Core/host pids and netns inodes, unit states, restart counts,
whether the reconciler had to act this boot, mount count, four non-mutating
probes, route ownership, and the full economy snapshot — plus the journal for
the boot so ordering is read from real timestamps rather than inferred from
unit dependencies.
It observes only; it never starts, stops or repairs anything, and carries no
Requires= or ordering that anything else waits on, so it cannot affect the
boot it is measuring. If the chain is broken the file says so, which is the
point.
Polling rather than a fixed sleep means a boot-time reconcile retry is
recorded as "settled late" rather than as a failure.
Two things surfaced by the production promotion.
`status` counted namespace mounts with an unanchored grep, so on production
"run/netns/openfut" also matched "openfut-staging" and reported a phantom
"2 = leaked stack" against a perfectly healthy host. A status command that
invents a fault is the same class of bug as a unit that reports active while
serving nobody, so it is fixed with an exact mount-point match. The bind and
reconcile logic is untouched; it always umounted an exact path.
openfut-rollback-detached.sh makes the documented rollback executable rather
than a paragraph in a runbook: it removes supervision, resolves the anchor's
CURRENT pid from Docker, and relaunches the incumbent detached pair with the
environment replayed from the captured env.json. --dry-run prints the exact
commands and touches nothing, which is how it was validated while production
was still being served by the processes it would restore.
Recreating the Docker anchor left the supervised Core and host stranded in
the dead namespace while systemd still reported them active — serving
nobody, invisible to any monitoring that trusts unit state. Reproduced on
staging: netns 4026539938 -> 4026540033, both pids unchanged in the old
one, both units "active", traffic ConnectionResetError.
There is no systemd-native edge signal to bind to. Containers do appear as
units, but the scope name embeds the container ID (docker-<id>.scope), which
changes on every recreate, so BindsTo= has no stable target;
NetworkNamespacePath= resolves once at start; a .path unit on /run/netns
would watch the file this tooling maintains. So: a level-triggered reconcile
on a 10s timer, comparing the namespace the services are ACTUALLY in against
the anchor's CURRENT one, acting only on a real difference. That cannot miss
an event while the watcher restarts or dockerd is down, and needs no
debounce — a burst of three recreations produced exactly one rebind. The
trigger stays separable: a docker-events unit could invoke the same script.
Anchor absent stops the dependants rather than falling back to host
networking; docker unavailable logs once and retries on the next tick.
Two defects found while testing and fixed here:
- mount --bind STACKS when the old mount is busy, silently leaking nsfs
entries; the bind helper now drains stale mounts in a loop.
- reconcile must stop -> rebind -> start, not rebind -> restart: a running
service holds the old namespace open and makes the umount fail busy.
Staging also gained a faithful anchor container so the reproduction is
structural rather than mocked. Economy state was byte-identical across every
lifecycle test. Production units are templates only and remain uninstalled.
Replaces the detached `setsid nohup … nsenter …` launch, which had no restart
policy, no boot persistence and no supervisor-visible logs. Staging units are
installed and proven; production units are TEMPLATES and are not installed.
Three decisions, each measured rather than assumed:
* `Wants=`, not `Requires=`, from host to Core. With `Requires`, stopping Core
stopped the host AND a later Core start did not bring it back -- a routine
Core restart would leave the client with no server. With `Wants` the host
survives a Core outage, answers 503 core_unavailable, never falls back to
Python, and resumes the moment Core returns with no intervention. Both halves
tested.
* Readiness is a bounded ExecStartPre TCP gate, because ordering proves nothing
about readiness and Type=exec only proves the binary exec'd. Core binds its
listener after migrations and content load, so "port open" is a real signal.
The gate FAILS rather than blocking: a host that waits forever looks healthy
while serving nobody.
* The netns is resolved by container NAME every start. The container is
restart=unless-stopped and its netns inode CHANGES on restart (measured:
4026539938 -> 4026540033), so a hardcoded pid is wrong by construction and
anything left in the old namespace serves nobody. Proven equivalent to today's
nsenter against a scratch container, never production's namespace.
`systemd-analyze verify` caught two real defects before deployment:
StartLimitIntervalSec/StartLimitBurst sat in [Service], where systemd 252
silently ignores them, so the crash-loop ceiling was not taking effect; and a
Documentation URL containing %20 parsed as a specifier. Both fixed and the
effective properties re-confirmed from the running units.
Staging evidence: Core-first ordering, host refused when Core is absent or
merely not listening, outage survival, automatic recovery, restart, graceful
stop with no strays, boot simulated via multi-user.target, 3x SIGKILL contained
at ~5s spacing, journald logs, and economy state byte-identical throughout
(integrity ok, fk 0).
ROOT CAUSE, one line. openfut-import-fifa17 emitted `"overall": 0` for every
non-player Core definition while `d.rating` already held EA's authoritative
`value` -- and the very next block wrote that same number correctly to the
adapter catalog. So the tier existed host-side but never reached Core:
Core overall 0 -> /collection effective_overall 0 -> CoreOwnedItem.rating 0 ->
tier_for_rating(0) = Bronze for a Gold (88) manager. That silent mis-grant is
exactly what the 409 was protecting against, so the refusal was correct.
The emitter now also writes `source_rating`, keeping `overall` at 0. Regenerating
the production pack changes exactly 18 entries and exactly one field each
(source_rating None -> value); same 1710 ids, same fingerprint 28c333f1e833338a.
WHY value IS the tier source, and why the thresholds are the player ladder:
LIVE_PROVEN, not inferred. The client re-rates staff from its own
managercards/*coachcards/physiocards by carddbid and applies discard_level's
65/75 ladder; coach_probe/discard_probe agree 4/4 (manager value 88 -> level 3,
coaches 66 -> level 2). The shipped coach tables corroborate: each family has
exactly 3 tiers x 2 rarities, and only 65/75 splits them 2/2/2.
Manager contracts stop refusing and now resolve the TARGET's tier from
Core-owned state. Still fail-closed everywhere it matters: a coach or physio is
`contract_target_not_a_manager` (only cardsubtypeid 4 is a manager), and a
manager Core carries no source_rating for is `manager_tier_unknown` rather than
a guessed tier. Core's own content_kind token is sent as target_kind, because
Core calls the squad manager `manager` while the catalog classifies it
`staff`+subtype 4.
NOT implemented, unchanged: STORED_MANAGER_BONUS and MATCH_CONTRACT_DECREMENT.
Promotion prep for the contract-apply cutover, which unlike the quick-sell
promotion moves BOTH binaries and applies a schema migration.
fifa17-promotion-backup.py uses SQLite's online backup API, not cp. Production
runs WAL with a routinely uncheckpointed WAL (515 KB at capture time); copying
the main file alone is not atomic against a live writer and carries no
guarantee the WAL holds no newer committed state. Emits a checksummed backup, a
metadata record and a RESTORE-*.sh that removes the stale -wal/-shm BEFORE
restoring -- omit that and SQLite replays the old journal over the file you just
put back, resurrecting the state you were abandoning.
fifa17-promotion-snapshot.py is read-only (mode=ro) and counts EVERY table
rather than a hand-picked list, so a delta cannot hide in a table nobody thought
to name. It also fingerprints the ownership rows, catching a row silently
rewritten when counts alone would match.
fifa17-contract-apply-validate.py gains --host/--db so one tool serves staging,
the migration rehearsal and the production acceptance run. Defaults stay
staging: there is deliberately no production default, so a bare invocation
cannot touch production.
POST /ut/game/fifa17/item/resource/<rid> {"apply":[{"id":N}]} now performs a
durable atomic contract application instead of falling through to Python.
THE RULE. grant = fcc_contractcards[card][tier(TARGET.rating)], then
min(99, contract + grant). The column is keyed on the TARGET's tier, NOT the
card's own -- all 36 cells of EA's shipped table match the published FIFA 17
matrix, and staging discriminates the two readings outright: a bronze-RARE
card on a rating-89 player granted 3 (the gold column), where the card-level
reading predicts 15.
No client binary reads fcc_contractcards -- a string scan of every .exe/.dll
in the install finds it referenced nowhere, and CardsDLL reads only 14 fcc_
tables (fcc_discardcoins among them, which is why quick-sell prices locally).
Consumable effects are server-authoritative, so EA's shipped table is the only
non-invented source and the client renders whatever we persist and re-serve.
The host computes the grant, Core owns the mutation -- the same split
quick-sell already uses (host prices via discard_value, Core performs
sell_item), and what migration 0027 means by "Core defines NO per-category
formula".
FAILS CLOSED, never 200-and-do-nothing: manager contracts 409 because staff
ratings are unimported so the target tier is unknowable; every other family
409 as unproven; batch 400; unresolvable operand 404. Core's deterministic
refusals pass through with their own status instead of collapsing to 503,
which would tell the client to retry a request that can never succeed.
`contract: 7` stops being a hardcode in shape_item/shape_staff_item and
becomes the fallback for an instance Core tracks no contract for. `fitness: 99`
is the same class of hardcode and is deliberately untouched.
CLEAN CUTOVER: Route::ConsumableApplyProbe, its handler, apply_probe_enabled,
the OPENFUT_FIFA17_APPLY_PROBE gate and both probe scripts are deleted. A
handler no classifier can reach is this repo's recurring defect class, and the
new economy arm preempts the probe. fifa17-migration-rehearse.py also drove
the probe (spelled "apply probe", so an apply-probe grep missed it) and would
have eaten a card off the rehearsal profile; retargeted to a non-mutating
assertion.
Not implemented, on purpose: the stored-manager bonus (real mechanic, rule
appears in no shipped table -- guessing it would corrupt the proven part) and
contract decrement per match (nothing spends contracts yet).
Fixing the display was only half of it. Quick-selling a consumable from the
repaired screen produced "There was a problem communicating with the FIFA
Ultimate Team servers", because the client's consumable quick-sell is a route
neither stack had ever served:
PUT /ut/game/fifa17/item/resource/5003068 body_len=0
Live-captured on staging. That path now carries three verbs -- GET is the
definition lookup, POST applies the consumable (ApplyCardByRes), PUT quick-sells
it -- and it is keyed by the stack's RESOURCE id, not an owned instance, unlike
the player quick-sell (DELETE item/<instanceId>).
This had to be Rust-owned rather than proxied: the Python oracle maps
item/resource method-agnostically to its definition route, so on production --
where the oracle is alive -- a PUT would return 200 with a definition list and
sell nothing, and the client would show a successful sale of a card the player
still owns.
Implementation reuses the retail-proven quick-sell path verbatim
(handle_quick_sell_path), so pricing comes from the same
ItemIdentityResolver::discard_value that stamps the number on the stack. Display
and payout are the same call; they cannot drift.
Two decisions, both documented in the code as decisions rather than discoveries:
* ONE copy per request. The request carries no quantity, and the screen prices
a CARD, so consuming a whole stack on one keypress would pay one card's
price for N cards. Selling one is the conservative reading.
* The copy sold is Core's first matching owned instance -- the same one whose
wire id the consumables screen already published as the stack's `item`, so
the player sells the card they were shown.
Verified against the running staging host:
displays 38 -> PUT -> coins +38, owned -1, consumables -1, stack 2 -> 1
replay sold the one remaining copy (+38, -1), no double credit
exhausted -> 404 not_owned, coins +0, owned +0 (no phantom payment)
123 host tests (+1 locking all three verbs on the shared path, and that the bare
`item` PUT stays the pile move), clippy -D warnings clean, fmt clean.
During the Rust production cutover four stale openfut-switch nft rules were
still redirecting production-facing traffic to staging (42127->42227,
8081->8281, 8094->18094, 8099->18106). They matched `ip daddr 10.10.0.120`, so
every server-side probe via 127.0.0.1 or the container IP passed while the
CLIENT was refused. That cost a full false-negative acceptance round: a retail
quick-sell landed on staging while production sat untouched, and the launcher
reported the server "not answering".
The failure mode is mechanical, so the check is:
* openfut-switch.sh status
* nft rules on OpenFUT ports, split into REDIRECT (interception) and DNAT
(docker publishing, expected -- reporting those as problems would train the
reader to ignore the tool)
* the actual point: loopback vs the ADVERTISED address per port. A redirect
keyed on the LAN IP is invisible to loopback, which is exactly why the
cutover probes all passed.
Verdict is CLEAN / INTERCEPTION_PRESENT with exit 0/1/2. Both branches
observed: it reports CLEAN now, and reported INTERCEPTION_PRESENT on a
loopback/advertised disagreement before :4216 was excluded.
:4216 is excluded from the verdict because LSX runs on the game machine --
compose publishes the port but OPENFUT_SERVERS omits lsx, so "published but not
served" is its normal state. It is still printed, marked as expected.
READ-ONLY by design: it never deletes a rule. Clearing interception stays a
deliberate operator act via `openfut-switch.sh off --name <id>`.
Run before production acceptance, client repoints, migrations and retail
protocol tests.
A production club displayed "Quick sell for 0 coins" for contract cards that
Core would have paid 3/13/32 for. The consumables stack wrapper hard-coded
discardValue (atom 0xd7) to 0.
The old rationale was that the client prices the card itself, the way it does
when we omit discardValue from an item. That is true of the ITEM record and not
of the STACK, and the evidence separates them cleanly:
* item+0x38 non-zero makes the client SKIP its local computation and show our
number -- re-proven on the live production client, 16/16 resident cards
"SERVER-SHOWN (local calc skipped)", including the acceptance card
235066 -> 40.
* We send no discardValue inside a consumable's item, so +0x38 is 0 and the
local computation DOES run and fills +0x3c correctly -- Milestone 1 measured
3/3/32/38 there, matching this table.
* The screen still showed 0. So the screen is not reading the item's computed
+0x3c; it reads the stack's atom 0xd7, which we were sending as 0.
So the number belongs on the stack, and it is the SAME
discard::value_for_definition that computes the payout -- one source, so the
screen and the wallet cannot disagree. Per CARD, not per stack: FUT prices a
card and the stack is only a quantity badge over identical copies. An
unpriceable definition stays 0 rather than inventing a number.
Verified on staging across every populated family, 16/16 stacks shown ==
recovered, none zero:
contracts 32/3/13 · healing 32/3 · training 3/13/34 · playstyle 38/38/38
· position 36/38/38/38/38
Two tests lock it: the payout equality (with the exact 3/13/32 the production
club would have been shortchanged on) and per-card-not-per-stack pricing for a
collapsed count=3 stack.
No payout logic changed, no taxonomy change, no ownership change, no Python.
250 adapter tests, 122 host, clippy -D warnings clean, fmt clean.
fifa17-migration-invariants.py Pre/post invariants across every domain the
migration authorization names: coins, ownership (+kind histogram, distinct
definitions, chemistry styles, loans, position overrides), squads,
managers, staff, consumables, club items, transfer state (market_listings),
packs, SBC, match history, plus integrity_check and foreign_key_check.
Table names are the REAL schema, not guessed: transfer state lives in
market_listings (29 rows in production), the FIFA17 opaque squad blob in
game_entity_ext.
fifa17-migration-rehearse.py Serves a migrated COPY with the candidate Rust
stack on isolated ports and validates the wire surface: club discardValue
is table-derived, squad projects, consumable categories populate, and the
apply probe is OFF (502 upstream-unavailable rather than a diagnostic ack).
Both are read-only against production: the rehearsal operates on a copy under
/home/alex/openfut-migration/, and nothing under openfut-promotion/state is
opened.
Evidence from the 2026-08-22 rehearsal is written up in the Vault runbook
"FIFA17 Rust Production Migration (rehearsed)".
Two corrections found while trying to close the effect boundary statically.
1. `contract: 7` IS OUR OWN PLACEHOLDER. fut_store.py:232's generic _item()
factory -- which builds every item the oracle serves -- hardcodes
playStyle 250 / contract 7 / fitness 99 on players and consumables alike. The
staging GK reads back exactly those three constants. So the production
catalog's contract:7 for resource 5001004 is an oracle placeholder
round-tripped through an observed profile, not an EA value. Its status is not
INFERRED, it is KNOWN-BOGUS as a source. Had the effect been implemented on
it, it would have been a fabricated game rule wearing observed-data clothing.
2. fcc_contractcards is NOT amount-less. An earlier note here claimed it "has no
amount column, so this value comes from observed data". It has 13 rows with
gold/silver/bronze/rating, 6 player + 6 manager paired by rating plus a
99/99/99 special. The sibling fcc_healingcards shares every column except
that it carries a single `amount`, which argues the differing columns ARE the
effect payload (per target tier). Against that: the values are non-monotonic
across tiers, which suits weights better than amounts; and no column of
5001004 is 7, so neither reading explains the placeholder.
The reader that would settle amount-vs-weight is in FIFA17.exe, not CardsDLL
(the table and column literals are absent from the DLL), so this stays
EFFECT_UNKNOWN rather than being guessed.
Also records, in content_taxonomy.rs, the competing reading of `development`:
fut_consumables.py's TYPE_CATEGORIES groups it as card-categories {6,7,8,9,10}
(modifiers only), explicitly flagged there as inferred from UI-bucket names and
never observed on the wire. Different enum space from the CONSUMABLE_TYPE switch
that actually emits the segment, and the switch gives formation/position/
playStyle/managerLeagueModifier their own segments rather than folding them into
development -- so the unfiltered reading is better supported, but it is still a
reading and the doc now says so instead of sounding settled.
248 adapter tests, fmt clean. No behaviour change.
Unit tests cover classification and body parsing; they do not prove the running
host behaves. These three scripts exercise the real service, and they found
nothing broken but make the two load-bearing claims checkable:
fifa17-apply-snapshot.py Core truth around an apply: coins, owned rows, kind
histogram, the source stack's copy count, and the
target's mutable fields (contract/fitness/playStyle/
training/injury). Coins and ownership come from the
staging DB, not the wire, so the check cannot be
satisfied by a projection bug.
fifa17-apply-probe-smoke.py Replays the EXACT captured request plus the edges,
against the live host, no client needed:
1. {"apply":[{"id":100000003}]} -> 200 {"itemData":[]}
source=Consumable subtype=201 copies=1,
target=fifa17_200389 rating=87
2. two targets -> 400 apply_batch_unsupported
3. unknown target wire id -> 200 UNRESOLVED_WIRE_ID
4. unowned source -> 200 NOT_OWNED
then re-snapshots: Core identical after all four.
fifa17-apply-gate-off.py The production-safety claim. With APPLY_PROBE unset
the same request must produce the pre-probe
behaviour, and does: no apply-probe line, three
passthrough lines, 502 into the dead upstream, Core
unchanged. Verified by restarting staging without the
flag -- an assertion about failing closed is worth
nothing unless the closed path is executed.
Nothing here writes to production; snapshot reads the staging DB read-only.
Claims POST ut/<sku>/item/resource/<resourceId> -- the consumable apply captured
live 2026-08-21 -- behind OPENFUT_FIFA17_APPLY_PROBE=1, default OFF. With the
gate off the route takes the extracted `passthrough` method, i.e. byte-for-byte
the behaviour that existed before this commit, so production cannot serve a
diagnostic even if the route is reached.
The handler is NON-AUTHORITATIVE BY CONSTRUCTION: it consumes no source card,
mutates no target, touches no contract/fitness/chemistry/training/injury state,
mints no coins and changes no ownership. It exists only to observe the client's
success path, because the EFFECT of a consumable is still unreversed and
implementing one on an inferred value is not acceptable.
RESPONSE SHAPE, from static RE rather than convenience (the brief was explicit
that `{}` must not be chosen because it is easy):
* The apply completion handler is CardsDLL 0x180035520. It does
`mov ecx,[rdx+0x1c]; test ecx,ecx; jne FAILURE`, raising
EVENT_CARDS_APPLY_CARD_SUCCESS (0x1801f37f0) on zero and
EVENT_CARDS_APPLY_CARD_FAILURE (0x1801f3810) otherwise. It tests exactly one
field -- the transport code -- and never inspects the body.
* That is materially different from the MOVE ack (0x180128600), which builds
per-item verdict records and reports FAILURE when the vector is EMPTY. The
`{}`-is-broken precedent does not transfer.
* The response object's constructor (0x1800a4ce0) initialises its record vector
(+0x50/+0x58/+0x60, 0x20-byte elements) EMPTY, so an empty parse result is a
legal state here, and the destructor (0x1800682b0) frees it accordingly.
* The legacy oracle routes `item/resource` method-agnostically to defs_route,
so historically this path answered with an `itemData` OBJECT.
`{"itemData":[]}` is the smallest candidate consistent with all four, and it is
labelled a PROBE, not a proven contract.
`apply` is an array, but only len==1 has ever been observed, so a multi-target
request is logged and refused (400 apply_batch_unsupported) rather than given
invented batch semantics.
Operands are identified READ-ONLY for the capture: the source by Core card id
(`<sku>_<resourceId>`, no new resolver method for a probe) with a copy count, the
target by reversing the wire id through the identity store -- never a guess,
`UNRESOLVED_WIRE_ID` when unknown.
Also records the reversed protocol and the `development` finding in
CLIENT_ROUTE_SURFACE.md.
122 host tests (+2: the verb/resource-id classification boundary, and target
parsing incl. the exact captured bytes). clippy and fmt clean.
The live client asked for `club/consumables/development` and got an empty
screen: `consumable_families_for_category` had no arm for it. Tracing that
segment recovered the client's OWN category vocabulary, and it is nine segments,
not the seven this file assumed.
CardsDLL, live 2026-08-22: the literal table at 0x1801f5a38 (under
MyClubAdapterClass / CONSUMABLE_TYPE) and the switch at 0x180048820, which
indexes by `enum + 1` through the byte table at 0x180048a90 into the case table
at 0x180048a6c:
enum -1 (unset) -> development
enum 1, 2 -> contracts
enum 3 -> healing
enum 4 -> fitness
enum 16 -> formation
enum 17 -> position
enum 23 -> playStyle
enum 24 -> managerLeagueModifier
enum 0, 5..15, 18..22 -> training (switch default)
Two consequences:
1. `formation` HAS a segment (enum 16). This file claimed the two formation
modifier families "have NO group code, so no segment can reach them -- that is
the client's own gap, not an omission here", and a test asserted it. Both were
wrong, and wrong in the direction that hides a server bug: it was our gap.
`formation` now maps to manager_formation_mod + formation_mod, so all
THIRTEEN families are reachable instead of eleven.
2. `development` is the type-UNSET bucket -- index 0 of a table indexed by
`enum + 1` -- i.e. no type filter. It is therefore the unfiltered view and
maps to every family via ALL_CONSUMABLE_FAMILIES. That is consistent rather
than overlapping by accident: the eight TYPED segments already reach all
thirteen families exactly once, so there is no family for `development` to
own privately.
The partition test now asserts the eight typed segments cover all thirteen
families with no duplicates, and that `development` is exactly their union, so a
family added to the taxonomy cannot silently vanish from the unfiltered screen.
Ownership and classification are untouched; this is projection only.
248 adapter tests, clippy and fmt clean.
Milestone 2: the consumable-apply protocol is now LIVE_PROVEN.
Adds opt-in passthrough BODY logging (OPENFUT_FIFA17_LOG_PASSTHROUGH_BODY=1,
default off, capped at 512 bytes) because a body is what names an unknown
mutation's operands, while also being the one place a request could carry
something that should not reach a log. Staging probe only.
With it, one operator apply captured the whole thing:
POST /ut/game/fifa17/item/resource/5001004
{"apply":[{"id":100000003}]}
source consumable : resource 5001004 (player contract, subtype 201) -- in the PATH
target item(s) : wire 100000003 (squad slot 0 GK, resourceId 200389) -- body apply[]
verb : POST
There is NO /apply endpoint, exactly as the static route work concluded. The
apply re-uses `ut/%s/item/resource`, which we already serve for GET (definition
lookup); the POST verb on that path is the mutation and nothing claimed it. This
is the wire form of the ApplyCardByRes task (id 0x0e), which is why the source is
a definition id rather than an instance id. `apply` is an array, so one resource
can name several targets.
Fail-closed verified: with the upstream dead the request 502s and Core is left
exactly unchanged -- coins 29,843,976, owned 1993, consumables 17, source card
still owned. No partial mutation.
NOT implemented: the response shape is unobserved and the EFFECT is unreversed.
Our catalog carries contract:7 for 5001004, documented as the matches granted,
but that is observed profile data (INFERRED), so no effect is written on it.
Bonus, caught by the same logging: the client really does request
`club/consumables/development`, which has no arm in
consumable_families_for_category and is served empty. Recorded, not guessed.
Host 120 lib tests, fmt clean.
Milestone 2 groundwork.
The passthrough arm forwarded to Python without ever recording WHAT was asked
for, so on staging -- where the upstream is deliberately dead -- an unhandled
request produced an anonymous 502. It now logs method, path and body length
before forwarding, which is how the next unclaimed route gets identified:
utas-host owner=PYTHON route=passthrough method=GET path=/ut/... body_len=0
Also records the FUT TASK vocabulary read out of the live client. The client
drives UTAS through named tasks held in a CardsDLL .rdata table of 0x20-byte
MixedCase/UPPERCASE slots, with a .data descriptor table giving each a task id:
ApplyCard 0x0d, ApplyCardByRes 0x0e, ConsumeCard, ActivateCard, AssingCard(sic),
MoveCard, MoveCardByRes, SwapCard, DiscardCard, DiscardCardByRes, ViewCards, ...
So consumable application IS a first-class client action even though the route
table contains no /apply endpoint -- it must ride an existing route. The
descriptor's function pointer is a `mov [rip+flag], cl; ret` setter, not a
request builder, so the request is assembled elsewhere keyed by task id; that is
cheaper to answer with one live capture than with more static tracing.
Search tooling carries mandatory positive controls (tradePile, ut/%s/item, squad
-- all FOUND), so the "no /apply route" result is a valid negative rather than a
failed scan.
Host 120 lib tests, fmt and clippy clean.
Adds scripts/fifa17-discard-impact.py (owned-instance economic impact, computed
from the shipped implementation's matrix -- informational, never a reason to
alter a value) and records the measured results.
Owned club, 1993 instances: legacy 1,820,700 -> recovered 19,128,031 = 10.51x.
Players 10.53x, manager 1.88x, consumables 0.19x (the ladder overpaid them ~5x),
staff 0.24x, club items 900 -> 0.
Adds two staging-only harnesses and records the results.
scripts/fifa17-discard-validate.py drives the REAL Rust/Core quick-sell path for
a fixture spanning every quick-sell-relevant category, and checks each against
the authoritative table value emitted by the discard_matrix example (i.e. the
shipped implementation, not a reimplementation). Per item it asserts the payout
is exact, the instance is removed exactly once, and a REPLAY of the same request
grants nothing and resurrects nothing.
Results with OPENFUT_FIFA17_DISCARD_TABLE=1 on the real 1993-item club:
players 6/6 exact 752 .. 74,400 (rareflag 1,3,4,5,6,11,21,22,23,24)
staff 2/2 exact 36 (gk coach, fitness coach)
consumables 4/4 exact 3, 3, 32, 38
club item 1/1 exact 0 (kit -- and 0 is what the client displays)
TOTAL 12/12 exact, 0 replay grants
wire discardValue == expected == actual payout for every player, so what the
client is shown and what Core credits are the same number by construction.
Concurrency: 4 simultaneous DELETEs on one wire id -> removed exactly 1, paid
exactly once (23,280).
scripts/fifa17-restart-persistence.py restarts Core and host IN PLACE with their
own environment rather than via the bring-up script, because `up` re-seeds the
club and would mask a persistence failure. It refuses to signal any process
outside the staging root -- production runs as another user and is skipped
explicitly. Result across SIGTERM + respawn of both: coins 29,967,428, owned
1978, players 1958 -> PERSISTED EXACTLY.
Production untouched; staging only, flag set only in staging.
`value_for_definition` declined for anything non-player without a catalog rating,
which sent club items into the legacy ladder and paid an invented 150 each.
That is wrong, and the client says so. `shape_club_item` sends neither `rating`
nor `discardValue`, and cardtype 7/9 are NOT re-rated by the client (the merge
jump table sends them to the shared tail), so the client computes for itself from
record +0xb4 == 0: level 1, `0 * price / 100` == 0. It DISPLAYS 0. Paying 150
invents value the player was never shown.
Move the decline boundary onto the real distinction, which is `client_rerates`:
* NOT re-rated (cardtypes 1, 6, 7, 8, 9) -> the server's rating is what the
client prices with, so Core's value is authoritative even at 0.
* RE-RATED (2, 3, 4, 5, 10 -- the staff families) -> the client substitutes its
own database value, so without a catalog rating we genuinely cannot match it
and must decline rather than guess.
Over the 1717-definition corpus this takes "declined -> legacy" from 6 to ZERO:
every definition is now priced by the one authoritative table and no generic
fallback is reachable in the current corpus. The six club items price at exactly
0; staff and consumables are unchanged.
Adapter 248 lib, host 120 lib, fmt and clippy clean.
The pricing DECISION (which rating to trust, when to decline) lived in the host
while the TABLE lived in the adapter, so FIFA semantics were split across two
crates and no single function could be pointed at as authoritative.
Move the decision into the adapter as `discard::value_for_definition(subtype,
rareflag, catalog_rating, core_rating) -> Option<i64>` and have the host call it.
Its three tests move with it. There is now exactly one table implementation, one
decision point (`ItemIdentityResolver::discard_value`), and one deliberately
retained rollback ladder (`legacy_discard_value`).
Add `examples/discard_matrix.rs`, which audits an entire FIFA17 corpus using the
SHIPPED implementation rather than reimplementing the formula, so the matrix
cannot drift from what the server pays. Over the current 1717-definition corpus:
declined -> legacy : 6 (badge, ball, kit x2, misc, stadium -- no catalog rating)
priced zero : 0
negative : 0
implausible : 0
rating boundaries : OK (1/2/3 at <65 / 65..74 / >=75)
Also re-verified both numeric cores against the LIVE client rather than trusting
the earlier notes:
level 0x180141e8a cmp al,0x4b -> 3 ; cmp al,0x41 ; sbb/add 2 -> 2 else 1
value 0x180141119 imul rating*price ; /100 via 0x51eb851f ; imul 0x64 ; sub ;
cmp remainder,0x32 ; jl/inc == round-half-up
`(rating*price + 50)/100` is identical to that for non-negative inputs.
Adapter 247 lib, host 120 lib, fmt and clippy clean.
Read the client's COMPLETE UTAS route surface out of CardsDLL's .rdata in the
running process (new tools/url_template_probe.py) and probed every one against
staging, where the Python upstream is deliberately dead so anything the Rust host
does not own answers 502 instead of being silently proxied.
That found five routes whose handlers already existed and were dead code because
`classify` never produced their Route -- the same defect as `season/list` and
`watchList`, whose fix comments are still in the file. This is the third and
fourth time:
captcha -> handle_static_ack, which already returns the oracle's exact
{encodedImg,sequence,sizeBeforeEncode}
tfa -> handle_static_ack, {}
livemessage -> handle_static_ack, {}
activeMessage -> handle_static_ack, {}
tournament/user-> FeatureOffEmpty, {} == the oracle with FUT_MODES off
(tools/utas_server.py:1504); the client builds this literal
at CardsDLL 0x18021e540 and the bare `tournament` arm never
matched it
Route's own doc comment already claimed the first four as "Rust-owned
UNCONDITIONAL", so the documentation was wrong rather than the intent. All five
are byte-identical to the oracle, so claiming them is parity, not new behaviour.
Invisible in production because the upstream answers there.
Two tests pin the vocabularies so a handler cannot go unreachable a fifth time;
both are mutation-checked (removing the captcha arm fails the first).
Also documents the surface in docs/CLIENT_ROUTE_SURFACE.md, including the trap
that bit me repeatedly: an .rdata literal is a FRAGMENT, not a callable path.
`clientdata`, `purchasegroup`, `sbs/challenges`, `squadBuildingSets`, `club/items`
and `item` all looked unserved and are not. Only `squad/mode` is genuinely
unserved, and correctly so -- it is Draft-only, which is out of scope.
L5 finding: there is NO consumable-apply route anywhere in the binary. The only
owned-item mutations the client can express are PUT item (move/pile), DELETE
item/<id> and POST delete/item (quick sell), and PUT squad. So applying a
consumable is not a dedicated endpoint; L5/L6 must be pursued by capturing the
PUT item payload, not by implementing a route that does not exist.
Host 123 lib + 45 host_test, fmt and clippy clean. tournament/user, livemessage
and activeMessage verified 200 on staging (were 502).
Serving owned balls (subtype 30), league logos (31) and fcc_misccards
(231/232/233/236) was the last projection gap. The open guess was that their
caption would come from `localizedName` on the wire, "probably", and they were
withheld out of caution.
Measured against the running client instead (new
tools/cardtype_dispatch_probe.py, read-only, reproducible, every step with a
positive control). They cannot be named at all:
1. The merge jump table at rva 0x141eb4 is indexed cardtype-1 with 10 entries.
Cardtypes 1..5 and 10 each get a DB-merge arm; cardtypes 6,7,8,9 ALL land on
one shared tail at 0x180141e8a that runs no query and writes no name.
2. `cmp [reg+0x4c], 9` (cardtype): ZERO sites in .text. For contrast, cardtype
1 has 13 and cardtype 7 has 6.
3. `cmp [reg+0x50], 30` and `..., 31` (cardsubtypeid -- the field that actually
selects a club-item caption): ZERO sites each, while kit 9, stadium 10 and
badge 11 all appear, which is the control. The only cardtype-9 subtypes
present anywhere are the four misccards ids, and all four are one boolean
predicate near 0x1801a72da that returns FALSE for them: an exclusion, not a
resolver. That predicate is NOT identified and is not claimed to be.
4. The cardtype-7 resolver is gated `cmp [rax+0x4c], 7` at 0x1800f6f04, so a
cardtype-9 item never reaches it. Its jne path formats AWARD_LABEL_%i --
the trophy path, not a fallback that would name a ball.
Nothing reads a localizedName for these subtypes, so sending one cannot become a
caption. Withholding them is a measured limit of the client, not caution, and no
server change can lift it.
CORRECTION: FUN_180119bd0 was recorded as "zero refs in CardsDLL -> almost
certainly an export, its caller is in FIFA17.exe". It is not an export. Its
address occurs exactly once in the whole process, at 0x18021c738 in CardsDLL's
own .rdata, and nothing in FIFA17.exe references it. It is virtual: vtable base
0x18021c2a0, slot +0x498, index 147 -- independently reproducing the recorded
"manager vtable slot +0x498" by a different method. Finding the boundary needs
the constructor-LEA trick; walking back over .text-pointing qwords runs 826 slots
through several adjacent vtables.
Bonus: the shared tail cardtypes 6-9 fall into IS the discard level ladder
(movzx [rdi+0xb4]; cmp 0x4b; cmp 0x41; store [rdi+0x54]), confirming
discard::discard_level instruction for instruction against the live client.
Adapter 244 tests, fmt clean.
The bring-up mints a manager (neither club owns one, and FIFA refuses to start a
match without one). Its catalog entry carried nation/league/team read out of
managercards but not `value` or `rare`, so discard pricing declined for it and
fell back to the placeholder ladder: 150 paid against the 282 the client computes
for itself.
Carry managercards.value 88 and managercards.rare 1, from the same table and with
the same provenance as the fields already there. `rare` is not cosmetic -- it
selects the discard price column, which is the whole difference between 282 and
97.
Both are live-confirmed on the running client: coach_probe grades the manager
record HIT (so +0xb4 == value and +0x58 == rare) and discard_probe reads the
value it computed for itself at +0x3c as 282.
Verified on staging: the manager now quick-sells for exactly 282. With this, every
card the client prices for itself -- manager, GK coach, both fitness coaches --
is paid the number it displays.
Staff quick-sell could not be priced correctly: for cardtypes 2/3/4/5/10 the
client overwrites the rating and rare flag we send with values from its own card
database, and a staff wire record carries no rating, no rareflag and no
discardValue at all. The server had no way to know the displayed price from what
it sent, so pricing declined for staff and fell back to the placeholder ladder.
The missing input was read straight out of the running client (pid 6580), no UI
interaction required:
* tools/coach_probe.py grades all four resident staff records HIT, which by
construction requires record +0xb4 == the table's `value` and +0x58 == its
`rare`. That settles `value`-is-the-rating, which was previously an inference
and was deliberately not shipped on that basis.
* tools/discard_probe.py (new) reads both discard slots -- +0x38, the value we
sent, and +0x3c, the value the client computed for itself:
1000509 sub 4 ct 2 rat 88 rare 1 sent 0 calc 282 predicted 282
9000081 sub 6 ct 10 rat 66 rare 0 sent 0 calc 36 predicted 36
3000083 sub 8 ct 4 rat 66 rare 0 sent 0 calc 36 predicted 36
4 of 4 agree, 0 disagree. 36 on the value-66 GK coach was the exact falsifier
written for this last commit.
Entities::enrich_staff now fills rating from `value` and rareflag from `rare` for
the five staff families, and the catalog emits the real rareflag instead of a
hardcoded 0 (it is not cosmetic -- it selects the discard price column, which is
why the rare-1 manager prices at 282 and a rare-0 coach at 36). Players and
consumables are untouched; their wire values are authoritative.
Verified on staging: a GK coach quick-sells for 36, not the 150 floor. The
catalog diff is exactly the two coach entries gaining rating 66; 1710 entries in
and out, nothing else changed.
The same probe shows what production does to PLAYERS today: every resident player
carries sent+38 = 1500, which suppresses the client's own computation, against a
real 688..752 for a gold rare and 72,800 / 74,400 for the two legends.
Still open, and not a discard problem: manager fifa17_1000509 is owned in Core
but has no catalog entry or definition (it reaches the client through the opaque
squad extension), so it declines to the ladder. That is definition coverage.
Importer 41 tests, fmt and clippy clean.
Quick-sell paid an invented five-tier rating ladder (its own comment said
"PLACEHOLDER, not EA-authentic"). It was blind to card type and rareflag, so a
94-rated TOTW special and a 94-rated gold common both sold for 1500, and every
non-player -- whose Core overall is 0 -- sold for the flat 150 floor. The ladder
existed in three places (adapter wire, host payout, an integration test's private
copy), which is a drift waiting to happen.
Add openfut-adapter-fifa17::fut::discard: the client's own fcc_discardcoins
table and its formula, round_half_up(rating * price / 100), keyed
(cardtype, level, rare). All of it is already reversed in
plan-2026-08-05-store-subsystem.md 3.6 and was verified there against 22 live
club items, 22 of 22 exact. DISCARD_COINS is generated from
fifa17-recon/data/tables/fcc_discardcoins.json and a test re-reads that file and
asserts row-for-row agreement, so the transcription cannot drift.
Collapse the three ladders into one method. ItemIdentityResolver::discard_value
both stamps the wire discardValue and prices the sale, because a non-zero
discardValue suppresses the client's local computation -- whatever is sent is
what the player is promised. The host's quick_sell_value is deleted and the
integration test's copy now calls the single implementation. A test with a
resolver double returning an impossible price proves the credit follows the wire;
reverting the payout to a ladder fails it.
Gated on OPENFUT_FIFA17_DISCARD_TABLE=1, default off: switching revalues the real
1991-item club 10.5x (1,820,400 -> 19,128,955 coins if wholly liquidated), up for
specials and DOWN for consumables, which the ladder overpaid 5.5x. That is an
operator's decision.
Staff decline to the ladder rather than pay 0: the client re-rates cardtypes
2/3/4/5/10 from its own DB and their rating is not imported. Deliberately not
guessed -- see the falsifier in the doc.
Verified on staging with the real club, both modes: flag off 1500 wire / 1500
paid; flag on 23760 wire / 23760 paid on an r99 rareflag-11 card (99*24000/100).
Consumables price from their catalog rating and agree with the client's own
computation. Adapter 244 lib tests, host 121 lib + 45 host_test, fmt and clippy
clean.
Decoded the vocabulary from the binary rather than trusting a case count:
FUN_18012ec50 is `cmp ecx,0x1d` plus a 30-entry jump table at 0x18012ed9c, each
case `mov ecx,<atom>; jmp <atom->string>`. Resolving those atoms against
fut_atoms.tsv yields the exact token list, and it matches club_type_filter
one-for-one — 30 implemented, none missing, none invented.
That is worth a test rather than a note. A MISSING arm answers a real tab with
unsupported_type and an empty screen; an INVENTED arm is worse, because it is
dead code that looks like coverage. Mutation-checked: renaming the leaguelogos
arm fails the test.
Two facts fall out that were previously guesswork. There is no
`playergoalkeeper` token — the client has only DEF/MID/FWD tabs — so a
goalkeeper appearing under playerdefender is CORRECT and not a filter bug, which
I had flagged as suspicious while sweeping. And `healing`/`contract`/`training`
exist as ?type= arms even though consumables have their own route.
Also completes the last unapplied item of plan section 7: the full vocabulary is
now written into ENDPOINT_MAP.md with how it was derived.
A health sweep of all 51 host routes found no errors, but did find a gap against
the documented club grammar: the client may send a comma-joined `defId=` list
INSTEAD of the filter block, and `parse_club_query` handled nine parameters
without it. Today such a request is answered with the whole filtered club rather
than the requested definitions — silently.
Deliberately NOT implementing the filter. That grammar is single-source (one
decompile plus one live log line, and the log line carried no defId), so the
reading of "definition id" is unconfirmed against any observed request. Narrowing
on a wrong reading would turn "too many items" into "zero items", which is the
worse failure and the harder one to diagnose.
So the parameter is parsed and reported instead: the filter summary gains
`defId=<n>` and the host logs a NOTICE naming the ids and saying plainly that the
response was not narrowed. The first real occurrence is then impossible to miss,
and the filter can be written against a captured request rather than a guess.
Verified live: the notice fires and total stays 1966.
"Place on Transfer List" is greyed for two reasons. This crate already fixes one
(owned copies emit `untradeable: false`). The other is `tradingEnabled`: the
client's struct defaults it to 0 — it is not a flag we have been overwriting, it
is a flag nobody has ever sent — and it gates the service half of the
TO_TRADE_PILE predicate (vtable slot +0x270, gate byte 0x1fd2e, measured 0 live).
`GET /settings` has always answered `{"configs": []}`.
The schema is high-confidence: FutGetSettingsServerResponse (deser 0x18013c6d0,
read end to end) has a single `configs` key holding `{type, value}` rows, and the
key ladder holds nothing else. `type` is the setting NAME. The row set is ported
from the shape the Python oracle would emit rather than invented.
Default OFF (`OPENFUT_FIFA17_COMMERCE_SETTINGS=1` opts in), because the flags are
RECOVERED BUT UNTESTED and the empty list is the live-proven body — the house
rule is that a flag defaults to the live-proven value. This also moves the
capability out of the oracle we are retiring and into Rust, where it can actually
be reached once Python is gone.
Verified against a real host on both settings: OFF returns {"configs":[]}
byte-identical to today, ON returns the 8-row body with tradingEnabled. It
explains why the menu entry is greyed; it does not promise the market works.
Rows 12, 13 and 16 carried guessed or placeholder URLs (`ut/%s/item?type=…`,
`ut/%s/…`). The real binding is a table, not an inference: the 125-row action
table at 0x1802caa20 indexes the 48-entry URL-base table at 0x18021df80 through
column 1, and base index 3 = ut/%s/club is carried by exactly four rows. So the
client can emit exactly four families on that base:
ClubSearch, ClubStats, StaffStats, ConsumablesSearch — which also corrects those
three rows to /club?<query>, /club/stats/staff and /club/consumables/<cat>, and
updates their status now that the Rust host serves them.
Added the complete club query grammar (ordered, with its suppression rules and
sub-vocabularies, including that the request spells it onSale where the response
says forSale), the seven /club/stats forms, the fact that /club/stats/team does
NOT exist, and the two base-table holes that are composed outside CardsDLL so
nobody re-derives them as findings.
Section 7 of the plan is now marked APPLIED and kept as the audit trail.
Section 7 of plan-2026-08-06-card-subsystem.md listed these and they were never
applied, so the stale text kept misleading readers — it already cost this project
a ten-row itemState table.
CARD_SYSTEM.md:
- "STILL UNKNOWN, AND NOT GUESSED" is ANSWERED. Its candidate set was wrong:
it asked which of 0x1e/0x1f/0x91..0x96 meant kit/badge/stadium, but three of
the five families are cardtype 7 (kit 9, stadium 10, badge 11) and are not in
that set at all, and 0x91..0x96 are trophies. Replaced with the settled map
and how each family's caption resolves.
- itemState table starts at 0x180229cc0, not 0x180229d20 — the recorded address
points MID-table, which is why six rows were missing. Added that
WAITING_FOR_GAME/inGame are aliases, that omitting the key yields invalid and
not free, and that the match is case-sensitive (measured).
- the consumables route claim "the /consumables/%s template ... the client has
still never used" is false; it IS that template, with base index 3 = ut/%s/club.
- added the dated field-map correction block, extended with the +0x60 and
definitionId findings measured on 2026-08-21.
tools/fut_store.py: the discard_value premise "that lookup returns no row for our
cards" / "WHY its lookup misses is still UNKNOWN" is false — it does not miss, the
tile reads a different property. That story sent one round of work chasing a table
defect that never existed.
Chased the league-logo lead to a useful boundary and stopped there.
FUN_180119bd0 — the cardtype-7 caption resolver the whole club-item story rests
on — has ZERO references anywhere in CardsDLL: no call, no jmp, address never
taken in .text/.rdata/.data. It is nonetheless a real function. An unreferenced
real function in a DLL is almost certainly an export, which puts its caller in
FIFA17.exe. So the owned cardtype-9 caption path is not in CardsDLL and looking
for it there is wasted effort; the launch probe remains far cheaper than parsing
the export table and 79 MB of EXE.
Also verified definitionId a fourth way, by a different method than the existing
three: every real atom name appears exactly once in CardsDLL's .rdata
(resourceId, cardsubtypeid, itemState, assetId, cardassetid, rareflag, owners,
contract, discardValue, and localizedName), while definitionId is absent
entirely. Recorded but NOT applied — the path carrying it is live-proven and the
saving is payload only.
Method note added: CardsDLL is .text 0x180001000, .rdata 0x1801e5000, .data
0x18028a000. Confusing a live mapping offset with an image offset reads the wrong
section and returns false negatives — it made every atom lookup, controls
included, come back ABSENT until corrected. Validate scans against a known key.
FUN_180098f20, previously described as the league-logo function with a hedged
"localizedName, probably", read in full: it queries fcc_leaguelogos WHERE
leagueid == %d, reads carddbid/value/cardassetid, and captions with
'LeagueName_Abbr_15_%d' in the 'FUT String' domain. A database-backed league
name therefore exists, in exactly the shape kits use for teamid — so "cardtype 9
has no DB name resolver" is too strong for league logos.
Deliberately NOT concluded: its only caller passes [rbx+0x20] as the league id,
and rbx there is a loop cursor over small list elements (int/double/int), not the
0x158-byte card record. Reading that as the record's assetId and shipping "send
leagueid as assetId" would be the exact inference this document exists to
prevent. Recorded as a lead with the next question named: does the OWNED render
path reach this resolver, and which field feeds it?
FUN_1801aa190 (the plan's "two minutes of work" item) is eleven instructions and
resolves TWO parallel arrays, not the one the earlier claim described:
f(self, idx, which) reads item+0x104+idx*4 when the flag is clear and
item+0x124+idx*4 when it is set — eight ints each, 0x20 apart. Live, BOTH read
all zeros on every resident record including a rating-94 player, so neither can
be the reason any action is greyed today.
The FUT roster database question is partly answered. Scanning FIFA17.exe in the
live process recovers the full API name set — StartFUTRosterDownload,
DL_FUT_LIVEDB, APPLY_FUT_LIVEDB, LoadFUTDatabase, UnLoadFUTDatabase,
SetFUTDatabaseUnloaded, UpdateFUTDBVersion, GetFUTDBCRC, RosterXMLDownloadedFail,
.dbFUTVer/.dbMajor/.dbMinor/CRCs — none of which exists in CardsDLL. That is a
downloaded, versioned, CRC-checked live database with its own lifecycle, which is
categorically not the shipped card tables. Whether it is loaded RIGHT NOW is
still open: the load flag was not located, and the absence of an open DB file
proves nothing since the process only holds Frostbite bundles.
The plan called this "the single blocker between 'we can mark a kit equipped'
and 'we can equip a kit'", and recorded that two attempts to find the writer
drowned at 1688 and 4144 instructions.
They drowned because +0x60 is a common struct offset. Two filters make it
readable: only an IMMEDIATE store can introduce a constant (a register store
just propagates one), and item-record code is recognisable by touching +0x4c
(cardtype) or +0x5c (itemState) within a few instructions.
Measured read-only against pid 6580:
- live +0x60 over all 27 resident records: {1: 23 players, 0: 4 staff}, never 4
- CardsDLL has 4 comparisons of +0x60 (0, 0, 1, 4); the 4 is the kit gate and
is the ONLY such comparison in the process
- CardsDLL has 29 immediate stores to +0x60, constants {-2,0,1,908,0x3f800000}
- FIFA17.exe, across 79 MB of code: ZERO stores of 4, zero comparisons with 4
- the gate function has one xref (a jmp) and its address is never taken
- every register store to +0x60 in CardsDLL is a struct copy or an init
So the gate is not a wire field we failed to send: the value it demands is never
produced by anything. Decoding it fully also shows every OTHER input is already
served — cardtype 7, itemState 101/102, teamid — leaving only the +0xba variant
selector beneath it, which makes a client-side patch the only remaining avenue.
Ball (30), league logo (31) and misc (231/232/233/236) were tracked as three
separate holes. They are one: cardtype 9 has no database name resolver, so the
displayed name can only come from `localizedName` on the wire, and that single
unproven step gates all three. The cardtype-7 families caption themselves from
the client's own tables, which is why kit, badge and stadium now project.
Ownership, content_kind, club/stats counting and restart durability are already
in place for all three, so the outstanding launch probe is the only remaining
work.
A club holding every ownable class, served through the REAL catalog resolver, so
each family travels the production classification path rather than a stub.
Asserts each family reaches its own `?type=` arm, that the staff arm carries the
manager too, and that `teamid` appears only where a caption resolves
TeamName_Abbr15 (badge yes, stadium no).
The cardtype-9 families are asserted WITHHELD with their catalog entries
RESOLVABLE, so an empty ball list is provably a decision about the family and
not an accident of a missing asset id — the two failure modes are otherwise
indistinguishable from the response.
Mutation-checked: reverting `is_cardtype7_club_item` to Kit-only fails this test
on the badge arm, so it guards the behaviour rather than merely describing it.
The plan recorded this as "almost certainly unresolvable statically", because
`FUN_180008190` is only a forwarding stub through a slot the host fills at
runtime: `mov rax,[DAT_1802ddfd8]; mov r9,[rax+0x248]; jmp r9`.
It IS resolvable — just not from disk. Read read-only out of the running client
(pid 6580): the slot forwards through two FIFA17.exe thunks into
msvcr120.dll+0x3c330, whose body is strncmp (`test r8,r8` count, `test al,al`
NUL stop, `cmp al,[rcx+rdx]`, then MSVC's 0x8080../0xfefe.. NUL-detect fast
path). No `or ..,0x20`, no folding table: the compare is raw bytes.
So the casing in the table at 0x180229cc0 is a CONTRACT. A mis-cased token does
not degrade gracefully — FUN_180166660 returns 0xffffffff, the record keeps 0 =
invalid, and the item fails the squad builder. This confirms what
fut::item_state already emits; it was previously true by convention and is now
true by measurement.
The probe follows the chain and attributes each hop to its module, which needs
care under Wine: PE sections are mapped anonymously, so a module is identified
by the nearest preceding named mapping rather than the containing one.
The guard classifies every definition id the game ships into exactly one of
KNOWN_OWNABLE / KNOWN_PRESENTATION_ONLY / KNOWN_UNSUPPORTED / UNKNOWN, and fails
when anything lands in UNKNOWN, so a table we cannot place is loud instead of
quietly assumed cosmetic. All 149 tables place: UNKNOWN = 0, 20,876 ownable ids.
Two tables are KNOWN_UNSUPPORTED with stated reasons rather than guessed at:
`fut_storymodehero` (80 ids; the shipped row is only {carddbid, teamid}, so no
item can be shaped without inventing one) and `fcc_misccards` (42 ids; these ARE
owned items, but their per-subtype semantics are not reverse-engineered).
It also counts by SHARED ID SPACE: `fcc_leaguelogos` and
`fcc_leaguelogostickers` both start at carddbid 8010000 and all 39 sticker ids
collide, so the union is 44 and never 83. The collision is printed so summing
cannot regress silently.
Staging now seeds the remaining club families (badge, ball, stadium, league
logo) from the client's own tables, so the rig exercises EVERY ownable class
instead of only the two the real club happens to hold.
Kit, badge and stadium are ONE record with ONE client-side resolver
(`FUN_180119bd0`, dispatched on `item+0x4c == 7`); they differ only in the field
their caption reads. Kits already ship and render, so the record is live-proven
— badges and stadiums were being withheld as if unreversed when the authority
(`plan-2026-08-06-card-subsystem.md`) marks both CONFIRMED, and its own rollout
order is "kits first, then badges, then stadia".
So the shaper generalises to the family, and carries exactly what each caption
resolves: `teamid` for kit and badge (`TeamName_Abbr15_<teamid>`), withheld for
stadium, whose resolver reads `StadiumName_<assetId>` and never looks at teamid.
Sending a field the resolver does not read is how this project earned a client
freeze.
Ball (30) and league logo (31) stay withheld. They are cardtype 9 with NO
database name resolver, so their name can only come from `localizedName`: the
offset is confirmed, but "the parser reads it" is not "sending it is safe".
Verified against the real club on staging: badge 6000005 emits cardsubtypeid 11
/ cardassetid 39 / teamid 21, stadium 6200000 emits cardsubtypeid 10 /
cardassetid 36 and no teamid, ball and logo emit nothing.
Kits were recognised by the id range 6_300_000..=6_400_654, so stadiums,
badges, balls and league logos all fell through to `Other` and were silently
dropped from the import — a club could own them and Core would never hear.
Club items are now settled by `cardsubtypeid` (kit 9, stadium 10, badge 11,
ball 30, league logo 31), which is the discriminator the client's own club-item
resolver uses and cannot collide with the other classes: consumables occupy
51..=341 and staff 4/6/8.
The render gate generalises with it. Each family ships a CONSTANT cardassetid
(kit 35, stadium 36, ball 37, badge 39, logo 40 — verified across all 2302
shipped rows), so a copy carrying anything else is deferred rather than drawn
as the wrong art. Only kits additionally require a teamid, because their
identity resolver keys on it and real owned kits carry it; demanding one of the
other families would defer every legitimate badge, ball and stadium.
League logos map to `misc`: they have no equipped slot, so Core holds them as
generic owned content rather than inventing a designation.
The real profile is unaffected (it owns no club items beyond its kits) and its
emitted catalog is byte-identical.
The snapshot predates the content taxonomy, so every fresh bring-up restored a
club whose coaches, kits and consumables were recorded as players. The wire
still looked right (the adapter classifies from its own catalog), which is
exactly the kind of divergence that hides until something keys off ownership.
Bring-up now runs Core's reclassify and asserts the club really owns something
of each kind. A Core binary predating the subcommand ignores it and boots the
server instead, so the step is bounded and says so rather than hanging.
The importer already knows every definition's kind, so it writes the mapping
Core needs to correct a club imported before the taxonomy existed. Applied to
the real 1989-item club: 20 rows corrected (17 consumables + 3 staff), 1966
players already right, 0 unmatched definitions.
Filling the bench writes squad rows behind Core's back, which invalidates the
stored FIFA17 opaque extension: Core saw a canonical squad that no longer
matched the extension's fingerprint, the host refused to apply it
(`stale_integrity`), and the client got a squad with zero players and no
manager. Nothing failed loudly — the rig just came up empty.
The seeder now recomputes the fingerprint exactly as
`services::squad::squad_fingerprint` does, and bring-up asserts the squad really
projects (>= 18 occupied, a manager present) instead of trusting that it did.
Seeded owned rows also state their content_kind, so Core does not record a kit
or a manager as a player.
Two defects that together made the club's 17 owned consumables invisible while
club/stats still counted them — the count gate promised 17, the item route
served 0.
1. The catalog omitted `card_asset_id`, `amount`, `contract` and `rating` for
non-player definitions. Without an art id the adapter refuses to emit the
card (it would draw the notfound box), and the families that read
`amount`/`contract` would render "-1" or grant nothing. All four values are
present in the source wire and were simply dropped on the way out.
2. Owned rows were imported without a `content_kind`, and Core defaults an
unstated row to `player` — durably recording a fitness coach and a contract
card as players in the ownership authority, even though the catalog-driven
wire looked right.
These are definition-level fields, so every owned copy must agree; a group that
disagrees is deferred rather than resolved by taking the first copy's value.
Measured on the real profile: observed `amount` equals the `fcc_*` table row for
every consumable that carries one (1, 2, 4, 5, 10, 15), and a wire omission
corresponds to a table amount of 0. It is therefore NOT a stack count — two
copies of 5003068 are two instances — so the import states no `quantity` at all.
Follows Core's kit designations becoming generic active-item slots: the host
reads `GET /club/active-items` (five always-present slots) instead of the
removed `/club/kits`.
Adds the consumables route and widens the club families to every content kind,
all resolved from Core ownership + the FIFA catalog. An item the client sees is
now an item Core actually owns.
Extends the FIFA17 adapter past players so the wire can carry the rest of a
real club's inventory.
itemState: the recovered 12-row table at 0x180229cc0 becomes the single source
(`fut::item_state`), replacing scattered literals. Every shaper draws from it
and the tests assert no shaper can emit a state the client does not know.
CARD_SYSTEM.md's 0x180229d20 is the middle of that table, not its start.
ContentKind covers all nine tokens. Managers stay inside the staff family for
counting, because the client's own club-stats model puts a manager INSIDE the
staff total with staffManager as a sub-bucket — a parallel Manager kind would
silently under-count.
Consumables get their own route (`club/consumables/<category>`) and a
stack-wrapper envelope, classified BEFORE the other club/ arms; they are not a
`?type=` family. This path previously fell through to Python, so owned
inventory was being served by the oracle.
The shaper refuses to emit a card it cannot render: no known art id, or a
missing `amount`/`contract` for the families that read them, or the subtype-219
rareflag trap that silently turns Player Fitness into Squad Fitness. A dropped
card is counted and logged, never faked.
Two things this project kept carrying as INFERRED are directly observable in the
card record, so this reads them instead of trusting the decompile:
rec+0x18 resourceId, rec+0x4c cardtype (derived by FUN_1800d8330),
rec+0x50 cardsubtypeid (as sent), rec+0x5c itemState (decoded enum value).
Measured against the live client (pid 6580, 27 records):
subtype 0 -> cardtype 1 (23 records) agrees with FUN_1800d8330
subtype 4 -> cardtype 2 (1) agrees
subtype 6 -> cardtype 10 (1) agrees
subtype 8 -> cardtype 4 (2) agrees
itemState runtime value 1 on all 27, and every one of those was served as "free"
So the cardtype map is now runtime-confirmed for every subtype we actually serve,
and `free == 1` is an empirical anchor for the itemState enum rather than a
reading of the table at 0x180229cc0. The probe prints the Ghidra prediction
beside each measurement and says DISAGREES rather than quietly matching, so it
stays useful as new families are served.
It also states the obvious limit in its own output: a runtime value only appears
if the client was actually served an item in that state, so absence is not
evidence of absence. The equipped states (activeBadge 100, activeHomeKit 101,
activeAwayKit 102, activeBall 103, activeStadium 104) remain table-recovered and
un-measured until a kit is fetched by the client.
Read-only: /proc/PID/mem is opened 'rb' and there is no write path.
`card_identity_probe` reads the PLAYER slots (F_NATION 0x148, F_LEAGUE 0x154). A
manager does not use those, so grading a manager with it reports nation=0 /
leagueId=0 and reads like a server bug when it is only the wrong offsets.
The manager layout, from the Ghidra reversal already recorded in fut_staff.py, is
teamid rec+0x94, nation rec+0xde, leagueId rec+0xe0, talkrating rec+0xe2,
negotiation rec+0xe3. The managercards merge (FUN_1801356c0) NEVER writes +0xde
or +0xe0, which is exactly what makes them a clean test: whatever sits there came
from our JSON and nowhere else.
Read against the live client (pid 6580, 27 records in the CardsDb map):
resource teamid nation league talkrating negot verdict
1000509 241 45 53 0 3 SERVER FIELDS LANDED
So the manager's chemistry fields DO reach the record, and `negotiation=3` agrees
with managercards row 1000509, i.e. the merge ran as well. That moves manager
nation/league from INFERRED to PROVEN without a screenshot.
Read-only: /proc/PID/mem is opened 'rb' and there is no write path.
FIFA refuses to kick off with "your squad must have at least 11 players and 7
subs … currently below the minimum number of players (18)". The imported club's
squad carries ONLY its starting XI, so a freshly installed real club is
unplayable until someone fills the bench by hand in the hub.
`fill_bench_to_minimum` tops the squad up with the club's best spare players,
writing EMPTY bench slots from index 11 upward. The 23 slots are 0..10 pitch and
11..22 bench/reserves, derived from the index alone, so the starting XI — and
any bench the operator has already chosen — is never touched and a re-run is a
no-op. Only real PLAYER definitions are eligible: a kit, a manager or a
consumable in a squad slot is nonsense the client would drop anyway. Selection
is best-rating-first with a stable id tiebreak, so the same bench comes back on
a re-run rather than shuffling.
Scoped to the REAL club on purpose. The 14-item fixture is a market test bed
with a single spare player; demanding 18 there would abort a bring-up that never
needed to kick off. Fixture mode therefore does not call this at all.
Verified against a copy of the club snapshot (the live staging database was left
alone, since it currently holds a squad the operator saved by hand): 11 -> 18
players, slots 0..17, no duplicate instance in two slots, every filled pick a
real player definition, and a second run filling nothing.
This is NOT a diagnosis of the failure the operator just hit — that squad had
already been filled to 23 valid players before the match was created, and the
host log shows the client issued no request at all after `match-create` beyond
account-sync, so that refusal is decided entirely client-side. It removes the
variable: after a clean bring-up the club is now playable without hand-editing.
"There was an error creating your game session. Please try again." on advancing
past the starting XI. The host log names it exactly:
utas-host ERROR passthrough to Python failed: … /ut/game/fifa17/match
utas-host owner=PYTHON_FALLBACK method=POST path=/ut/game/fifa17/match status=502
Neither `match` nor `match/end` was claimed by either classifier, so both fell to
Passthrough. This is the third instance of one defect: `season/list` and
`watchList` were the first two, and like `watchList` the handler already existed
and was simply unreachable — `EconomyRoute::MatchEnd` was produced ONLY by
`POST /ut/delete/game/<sku>/match`, a URL the retail client never sends. The
adapter's `match_wire::create_response` had zero callers.
The family is now classified by PATH SUFFIX and is deliberately VERB-AGNOSTIC:
the strings "PUT" and "DELETE" do not occur anywhere in cardsdll.dll, so verb
selection happens outside the DLL and cannot be pinned statically. Matching on a
verb is precisely how these came to be proxied. All three arms live in the
ECONOMY classifier, because `/match/end` credits coins and `try_handle_economy`
is the barrier guaranteeing a claimed route can never also reach Python — and
because create and end must share the in-flight match id, splitting the family
across two classifiers is what let them diverge.
`POST …/match` is both FutCreateMatch and FutPlayGame, discriminated by an
integer `matchId` in the body exactly as the client serializes them; play acks
`{}` and must not mint a second session. `squad` is omitted from the create
response: nested, half-read, the documented freeze mode.
MATCH IDS GET THEIR OWN IDENTITY SCOPE. The oracle mints them from the same
counter as owned items, which is why an observed match id looks like an item id,
but that is an artifact of a single-counter save file. Here the identity store
keeps a real reverse map, so an item-scoped match id would make
`owned_id_for_wire` resolve a match to a bogus owned card and corrupt quick-sell
and move. A new `(game, "match")` scope costs one constant — the store is
already generic over the pair — and an integration assertion now pins that a
match id never appears in the owned-item reverse map.
ECONOMY: `/match/end` is NOT a new authority. It renders Core's single
exactly-once `complete_match` transaction, the same one the legacy
`/matches/result` path was closed in favour of earlier today, and it still omits
`expire_loans`/`advance_season` so FIFA 17 keeps its own seasons and loans.
THE LATENT BUG THIS EXPOSED, which would have been a silent permanent
under-credit the moment the route became reachable: the per-match identity fell
back to a hash of the request body. Every abandoned match sends a BYTE-IDENTICAL
body (`matchReportId:0`, empty items/matchData/telemetry, flags 0), so all of
them collapsed onto one identity and Core's UNIQUE(profile_id, match_identity)
would refuse every DNF after the first — `applied=false`, nothing awarded, no
error. The identity is now the id minted at create, which is unique per match by
construction; the fingerprint remains only as a floor for an end with no create.
It also removes a durable dependency on `DefaultHasher`, which has no
cross-version stability guarantee yet was being persisted.
One bug of my own, caught by driving the real dispatch rather than the handler:
taking the in-flight id on end looked tidy but sent a REPLAYED `/match/end` down
the fingerprint path — a different identity — so Core paid a second time
(measured: a second +75 for one abandoned match). The id is now read and held,
so a replay reuses one identity and the next create overwrites it.
Verified end to end on the restored club: create → ready → play → end returns the
reversed reward shape (`boostConis` included, `bidTokens`/`qualifiedChampionEventId`
never emitted), a DNF credits once, two replays credit zero, and a second match
with a byte-identical body credits again. Host 115 lib + 36 host_test + 7
economy_integration + concurrency/differential/failure, adapter 217 + 25, all green.
Note for the record: a DNF pays Core's COINS_LOSS (75), not the oracle's 100.
Nothing on the wire settles the number — the client renders whatever we send, and
the oracle's own comment says its values were never reversed — so the declared
Rust authority's table wins rather than being bent to match Python.
The operator picked a manager in the FUT hub, then found no manager on the
pre-match squad. The save was NOT the problem: the host logged three
`route=squad-replace status=200 outcome=ok detail=[]` with no unresolved ref,
Core wrote the `squad_managers` row, and every read projected the assignment
back. The client simply had nothing to draw.
`squad.manager[]` was emitted as a bare `[{id, dream}]`. That looked
retail-faithful, and the previous commit defended it on the grounds that no
capture had ever shown otherwise. Re-reading the captures with a populated
manager in hand shows why that was the wrong conclusion: every retail capture
carrying the bare form has `id: 0` — an EMPTY manager. None of them ever
demonstrated that a POPULATED ref renders without its item, because none of them
had one. `plan-2026-08-05-families.md` says as much outright: "FUN_18013d1f0 was
never read for a staff member".
The squad object is self-contained everywhere else: `players[].itemData` carries
the whole card rather than an id resolved out of band. The manager is the same
kind of slot in the same object, and the one implementation that ever drove a
working manager — the Python oracle's squad — emits `id` BESIDE `itemData`.
The element shapes differ and both are now pinned by tests: a player slot is
`{index, itemData, kitNumber}`, the manager is `{id, itemData, dream}`.
So the manager is projected through `resolve_staff` and its item embedded with
`shape_staff_item`, the same 11-key record `/club` serves. An assignment with no
resolvable staff identity still yields `[]` rather than a fabricated ref.
`STAFF_CONTRACT` moves next to `shape_staff_item` in `fut::item` (re-exported
from `club_response`) so `/club` and the squad cannot disagree about the
contract the client checks before kickoff.
Verified on the restored club: userMassInfo, /squad/0 and /squad/active all
carry the manager with resourceId 1000509, contract 7 and the nation/league/team
the client cannot supply itself. Adapter 217 lib + 25 integration, host 114 lib +
36 host_test and every economy suite green.
FIFA refuses to kick off with "your player or managers contracts have expired".
The club had no manager, and could not have had one: `/club?type=manager` (the
token the STAFF tab actually sends) was rejected by the host, and staff items
were counted and dropped by the adapter instead of being shaped.
The squad's manager reference is a red herring worth recording. It points at
wire id 100000427, which resolves to resourceId 3000083 = a FITNESS COACH
(cardsubtypeid 8), not a manager. The client's own club/stats agrees:
staff:3, staffManager:0, staffGKCoach:1, staffFitnessCoach:2. This club has
never owned a manager, so one is MINTED rather than restored.
Wire shape is not guessed. `fifa17-recon/tools/fut_staff.py` is an
instruction-level reversal of the item parser and the managercards merge that
justifies every key by its record offset, and CARD_SYSTEM.md records it
confirmed live on 2026-08-05 (ten managers rendered with correct flags, league
names and "CONTRACT 7" on the card front). `shape_staff_item` emits exactly that
key set and nothing else:
* `nation` (rec+0xde) and `leagueId` (rec+0xe0) are MANAGER-ONLY slots the
client's merge never writes, so the server is their only source — they are the
flag, the league badge and both halves of manager chemistry. Coaches get
neither, because the four coach tables have no nation/league/team column and
emitting zeroes there would be invention.
* `resourceId` is the RAW merge key: staff are read as a u32 with NO &0xffffff
mask (players are the only masked family), so `version` must stay 0 or the
lookup misses — silently, since the manager branch has no else-arm.
* `preferredPosition`/`attributeList` are omitted because they SURVIVE the merge
and are then read by the card view-model; `assetId`/`rating`/`rareflag` are
omitted because the merge overwrites them from the client's own tables. A
staff card is therefore never routed through `shape_item`.
Managers stay inside `ContentKind::Staff`, discriminated by `cardsubtypeid == 4`
— the client's own discriminator, and its own stats model counts a manager
INSIDE the staff total with staffManager as a bucket within it. A parallel
`ContentKind::Manager` would have been a second source of truth for a fact the
subtype already carries, and would have silently under-counted club/stats.
`squad.manager[]` stays `[{id, dream}]`. The only populated form anywhere is the
oracle's DRAFT squad; no capture has ever shown itemData in a regular squad, and
feeding that deserializer the wrong container type freezes the SAX reader. The
contract reaches the client through the CardsDb record registered from the
/club envelope, which is a find-or-insert and therefore accumulates.
TWO SILENT BUGS FOUND ON THE WAY, both of which made a correct assignment look
like no assignment at all:
1. `get_squad_manager` read `manager.owned_card_id`, but Core returns the
assigned OWNED CARD, whose field is `id`. It therefore ALWAYS returned None —
indistinguishable from "no manager". Now reads `id`, and a present-but-
unreadable manager is an error rather than a silent absence. The projection
also now warns when an assignment cannot be resolved to an owned instance,
which is the documented "Core drops an owned card with no CardDefinition from
/collection without erroring" trap.
2. `Route::WatchList` was produced by NO classifier arm, so its handler was
unreachable and every `watchList` request fell through to Passthrough — the
same defect class as `season/list`. Against a stack whose Python upstream is
deliberately dead this 502'd. This was failing
`sbc_survives_complete_core_and_host_restart` at HEAD before this change.
The manager itself is seeded from the client's own tables, never invented:
managercards 1000509 (assetid == carddbid), nation 45, manager[509] "Luis
Enrique" teamid 241, leagueteamlinks 241 -> league 53. League 53 is also the
dominant league in the restored squad (12 of 23), so the chemistry pairing is
the correct one rather than an arbitrary pick.
Verified live against the restored club: /club?type=manager and ?type=staff both
return 4 items (the minted manager plus the 3 coaches the profile already owned
and could never see), the manager carries contract 7 with nation/league/team,
coaches correctly carry none of the three, squad.manager resolves to the same
wire id, and no staff leaks into ?type=player. Adapter 219 tests, host 114 lib +
36 host_test + all economy suites green.
The operator could not field a starting XI because staging has only ever held the
14-item synthetic fixture (11 auto-picked starters, one disposable, two kits). The
real club -- the 1986-item CAGE import, 29,843,976 coins -- was never lost, but it
sits in `/home/alex/openfut-promotion/state`, which BOTH staging lifecycle scripts
list in FORBIDDEN_PATHS and refuse to open. That guard is correct and stays.
Worth recording while looking for the club: the LIVE production Core container
serves an EMPTY database (0 owned cards, schema predating even the game_id column).
The real club is not being served anywhere right now; it exists as state on disk.
So restoring it into staging is not a convenience, it is the only way to play it.
Two pieces:
`scripts/club-snapshot.py` is the ONE place allowed to read production state, and
it is read-only by construction: the Core database is opened `mode=ro` and copied
with sqlite's online backup API (a plain file copy can tear a database with a hot
WAL), every destination is asserted to be outside the production directory before
anything is opened for writing, and the sha256 of every source is compared before
and after -- a mismatch aborts, because that would mean the snapshot modified
production. It then proves the copy is faithful (same counts, coins, squad) and
that the identity store maps EVERY owned card to a wire id, since an unmapped card
would reappear under a freshly minted id and break the client's cached squad.
`sold-staging-up.py --club real` installs that snapshot. It is installed BEFORE
Core first starts, so Core migrates the copy forward from schema v19 through
match_completions, squad managers and kit assignments. Seller A is then already
present -- it IS the imported persona -- so only Buyer B is seeded, the kit
fixtures are attached to the real club so the kit work stays exercisable, and the
real squad is left alone. The up script still never reads production state: the
snapshot lives outside it, which is precisely what makes `--club real` compatible
with the `safe_path()` refusal.
The resolvability preflight now covers whichever club will actually be served. This
is the check that matters most for the real one: Core does not fail on an owned card
whose definition is missing, it silently filter_map-drops it, so a gap shows up as
an EMPTY club with all 1986 rows still in the database. Verified: all 1712 distinct
card ids resolve in both the content pack and the identity catalog, 0 missing.
`sold-staging-seed-squad.py` now REFUSES to run when the manifest says the real club
is installed. `PUT /squad/0` is a full replacement, so the fixture seeder would have
overwritten the operator's own lineup with an auto-picked XI -- destructive and not
recoverable in place. `--show` still works in every mode; `--force` overrides.
Verified end to end against the restored club: /club 29,843,976 coins, /collection
1988, 1966 players + 2 kits served over the UTAS wire, squad 'OpenFUT' (f433) rated
90 with 11 players carrying contract 7 / fitness 99, and every wire id stable from
the snapshot identity store. The fixture path was re-run afterwards and still seeds
exactly 14 items, so the SOLD experiment is unaffected.
Single-player Seasons failed with "There was a problem communicating with the
FIFA Ultimate Team Servers". Two independent faults, both fixed:
1. The client never reached a season endpoint at all. It aborts on a
prerequisite web file, captured live by the deployed trace:
SEASONS_WEBFILE_URL: url="packs/loc/storepackdescriptions.en_us.xml"
SEASONS_STAGE1: status(+0x1c)=999 -> CACHE_PACKNAMES_FAILED
That is the hook's side (launcher 5294f58): the CDN base is empty in the
emulator, so the url stays relative and never reaches the POW content server
on 8085 that actually serves it.
2. `season/list` and `season/user` were not served. Only the EXACT tail
"season" was classified (as FeatureOffEmpty); every sub-path fell through to
Passthrough — the deliberately-dead Python upstream — so the mode could not
have worked even once the web file resolved.
Adds `fut::season_wire` with the reversed element schema (parser FUN_180167740,
stride 0x318; matches elements via FUN_180167fb0) and a `Route::Season` owning
`season…` for GET plus the state-storing PUT. Anything else still proxies rather
than being claimed without evidence.
Two things the types encode because getting them wrong is fatal:
* `matches` is NEVER empty. StartSeason (FUN_1800fc500) indexes
`matches[*(x+0x70)].teamId` off `elem+0x2e8`; an empty vector makes that a
NULL dereference and the client dies at CardsDLL+0xfc5b5. A full ten-round
schedule is emitted, with opponents drawn from team ids observed in this
client's own database.
* `type` MUST serialise before `divisionId`. `serde_json::Value` is a BTreeMap
here (no preserve_order), so `json!` sorts keys ALPHABETICALLY and emitted
divisionId first — caught by a test written for exactly this. The wire shapes
are therefore `#[derive(Serialize)]` structs (declaration order) rendered
straight to text via a new `json_text_status`, never round-tripped through
Value.
Verified on staging: season/list returns the ten-round OFFLINE season with
type before divisionId, season/user the round-1 position, history an empty
list, and the bare tail still {}.
Season progress is not yet persisted: the state-storing PUT is acknowledged
with {} (what the retail wire answers) rather than pretending a round advanced.
Core bae0a2b: `POST /matches/result` fails closed (it was a second economy
authority with no transaction and no idempotency key); loan expiry and Core
season progression move into `complete_match`'s transaction behind opt-in flags
that default OFF, so the FIFA 17 retail path is unchanged; notifications emit
post-commit and only when applied; `/auth/reset` now clears `match_completions`,
which previously made any profile that completed a match unresettable.
Bridge 07e83fe: the two EA result routes and the dashboard submit to
`/matches/complete` with a per-submission `match_identity`. Pushed to
`fix/matches-complete-migration` — the bridge pin is deliberately behind its
origin/main, so this does not touch that branch.
Every real squad save was failing with 400 unresolved_wire_ids. Reproduced on
staging with the repo's own seeder, which sends the captured retail body:
route=squad-replace status=400 outcome=unresolved_wire_ids detail=[[100000427]]
FIFA 17 always sends a manager ref, and on a real profile it does not resolve to
an owned instance. scripts/sold-staging-seed-squad.py already recorded why:
production's own squad points at instance 100000427, which is absent from
production's /club/staff (1975 items spanning 100000001..100004826), and the
client accepts that squad back unchanged -- so the client never validates the
manager against the club, and the pre-0023 server accepted it.
Making the manager ownership-backed (d37a9d5 / 25f4ad1) turned that ref into a
hard refusal, which took out the primary FUT write path: no squad save means no
squad, which means the client will not enter the FUT hub at all.
A manager ref is not a squad slot. An unresolvable PLAYER slot must still refuse
the save -- committing it would silently drop an owned card from the club. An
unresolvable MANAGER ref just means there is no ownership-backed manager, which
is exactly the state before migration 0023: the save commits, the assignment is
cleared as a full replacement should, and the id is reported on
ProposedSquad::unresolved_manager_wire_id so the host can log what it could not
map instead of letting it vanish. A ref that DOES resolve is still assigned and
still authorized against the club.
Verified end to end on staging: the seeder now answers {"id": 0} with 11
occupied slots, the host logs
`manager_ref_unresolved=100000427 (saved with no manager assignment)`, and the
projected squad carries `manager: []`.
Staging served `kits=2 kitsHome=0 kitsAway=0`: the split compared the catalog
`asset_id` against `fcc_kitcards.assetid` (14/15), but a kit's catalog
`asset_id` IS its carddbid (6300006), not that column, so neither family ever
matched.
The carddbid range is the same fact in the form we actually carry: across all
1482 kit rows, assetid 14 covers precisely the 828 `63xxxxx` ids and assetid 15
precisely the 654 `64xxxxx` ids, with no exceptions either way. Keying on the id
we already have avoids carrying `assetid` as a second source of truth for the
same split. Tests now use the real team-21 pair (6300006 home / 6400003 away).
Verified against the staging stack: `kits=2 kitsHome=1 kitsAway=1`, team-21
bucket 2, `?type=kit` still activeHomeKit/activeAwayKit, `?type=player` 12.
Four defects found by running the suites and the staging lifecycle end to end
after the kit milestone.
1. club-stats kits were half-implemented. The global `kits` counter was real
but `kitsHome`/`kitsAway` and every per-team `kits` bucket stayed hardcoded
0, so the same screen reported two owned kits and zero home/away kits.
`kits` is a total with a family split, exactly like players/playersGold and
staff/staffManager. The split key is `fcc_kitcards.assetid`: 14 is the home
family and 15 the away family, verified across all 1482 rows of the kit
table (assetid 14 covers exactly the 63xxxxx carddbids, 828 rows; assetid 15
exactly the 64xxxxx ones, 654 rows; no exceptions either way).
ClubStatInput now carries `asset_id`, and a kit buckets onto the team that
wears it -- including a team the club owns no player from, the normal case
for a kit won from a pack. The host reads both from the catalog through new
NON-MINTING accessors: `resolve`/`resolve_kit` allocate a wire id, which a
read-only stats query must never do as a side effect.
2. host_test.rs had 10 tests red since the squad-manager work (25f4ad1 /
d37a9d5); 56bd9dd updated the squad_projection integration test and stopped
there. `put_body` hardcoded the captured manager ref 100000427 into EVERY
save, including tests with no manager fixture, so each one was refused with
`unresolved_wire_ids` -- the tests were reporting a real invariant against a
fixture that could not satisfy it. The manager is now an explicit
`Option<i64>` per test, and FakeCore models Core's manager persistence
instead of inheriting the "not implemented" default that 502'd every save.
Added the coverage whose absence let this rot: a manager assignment
round-trips as a Core owned id, a later save without one CLEARS it, and an
unowned manager ref refuses the whole save with nothing committed.
3. `club_route_maps_query_and_shapes_core_items` pinned `offset`/`limit`
forwarding to Core, which the kit commit deliberately replaced with
host-side pagination. It only ever passed because FakeCore ignored the
window -- against a real Core, `start=10` over a one-item club was always an
empty page. Retargeted to the real contract (Core gets semantic filters and
NO window) plus a new test that the window is applied locally after
filtering, which the old fake made vacuous.
4. The staging lifecycle scripts identified production by hardcoded pids, so a
correct teardown FATAL'd: production moved into containers and pids
3631953/3374264 died with a container restart days ago. A pinned pid rots
into the worst of both worlds -- a kill-refusal gate that no longer names
any real production process, and a liveness gate that fails a healthy
teardown. New shared `scripts/openfut_production.py` resolves production
pids AND published ports from the container runtime at the moment they are
needed, refuses to signal anything it cannot see, and proves production is
the same processes serving the same ports before and after. Both lifecycle
scripts use it, which also closed a real gap: port 8085 is published by
openfut-fut-backend but was missing from the up script's forbidden list, so
staging could have bound a production port.
Also fixes the economy differential, red because `complete_match` unlocks
achievements in the same transaction that pays the match reward -- a deliberate
Core feature the Python oracle has no counterpart for. `rust WIN +400` asserted
that progression did not exist; it now asserts the delta is the 400 match reward
plus exactly the achievements the match unlocked, read from Core's own report.
Closes the server side of the FUT kit selector. Ownership stays generic in
Core (submodule bump: club_kit_assignments + GET/PUT /club/kits); this
commit adds the FIFA17 representation, the host projection and importer
support.
adapter:
* ContentKind::Kit ("kit") so kits are classified alongside player/staff/
consumable instead of being mistaken for 0-rated players.
* Fifa17CardIdentity carries card_asset_id and team_id; RawCard keeps both
optional because the emitted catalog writes null for non-kit definitions.
* shape_kit_item emits only the fields the client's kit path reads
(id/resourceId/assetId/cardassetid/cardsubtypeid/itemState/owners/
untradeable/teamid) — no attributeList, no itemType.
* itemState on the wire is the STRING token activeHomeKit/activeAwayKit;
the 101/102 integers are the client's post-deserialisation runtime enum
(item+0x5c) and are never emitted.
* club_stats S_KITS (0x28) now counts owned kits instead of a hard zero.
host:
* CoreKitAssignments + CoreAccess::get_active_kits (GET /club/kits),
defaulting to no active kits so a Core without the endpoint degrades
instead of fabricating a designation.
* handle_club classifies type=player|kit, rejects any other type with an
empty page and outcome=unsupported_type, and now always fetches
unpaginated from Core: kind and transfer-pile membership are host-side
concepts Core cannot express, so filtering and pagination must both
happen after shaping or pages come back short.
importer:
* ItemClass::Kit (cardsubtypeid == 9 and resourceId in 6_300_000..=6_400_654),
kit counts/balances, and card_asset_id/team_id carried into the emitted
catalog and manifest.
* a kit group missing cardassetid == 35 or teamid is DEFERRED
(missing_kit_render_metadata) rather than defaulted; conflicting render
metadata across instances defers as render_metadata_conflict.
staging: sold-staging-up.py seeds two owned kits (6300006 home / 6400003
away, team 21) plus both active designations so the projection can be
verified over HTTP before involving the client.
Points at launcher 00ad631: retire FIFA 23 as a build target/template while
keeping the hook game-generic by per-game feature, plus the earlier config-driven
FIFA17 socket redirect (16f3452). Build invariant --features fifa17 unchanged.
Add read-only preflight verifier and Windows-client documentation for the
reimaged native-Windows FIFA17 client host (10.10.0.105). No launcher script:
the native model is _fifa17.exe run as admin (RUNASADMIN + shortcut). Covers
routing (openfut.cfg -> 10.10.0.120), rollback (version.dll swap), and the
x64dbg RVA<->VA (ASLR) attach workflow (ImageBase 0x180000000 CardsDLL/powdll).
The manager moved from an opaque extension field to an ownership-backed
canonical assignment: SquadProjectionInput.manager (owned item) replaces
Fifa17SquadExtensionV1.manager.
- project_put stand-in passes manager: None; baseline asserts the manager
projects empty when none is owned.
- persisted_read registers the manager's owned instance + identity and
passes it as the assignment, asserting the resolved [{id,dream}] ref
round-trips to the read oracle's manager.
Adapter: new fut/match_wire.rs owns the FIFA17 match wire — endReason
enum -> canonical result token (win/draw/loss/dnf/no_contest), match-end
payload parse (goals from myMatchStats[0], omitted on DNF/QUIT), and the
reward-response projection (only reversed fields; never bidTokens/
qualifiedChampionEventId). Match logic removed from economy_policy.rs
(kept pack/fee); registered match_wire in fut/mod.rs.
Host: handle_match_end now applies the match to Core's authoritative
complete_match (POST /matches/complete) fail-closed — any Core error is a
503, never a Python fallback — and renders Core's authoritative coins.
Per-match identity from matchReportId or a body fingerprint keys Core's
durable idempotency. New CoreEconomy::complete_match transport +
CoreMatchCompletion/CoreMatchReceipt.
Tests: adapter endReason/parse/projection; host shaping, fail-closed,
malformed, identity dedupe; integration replay + rebased balance chains
(match now also grants XP/level-up/achievement coins).
Thread the manager assignment between the FIFA squad path and Core:
- CoreAccess gains get_squad_manager/set_squad_manager (GET/PUT
/club/manager); HttpCoreClient implements both.
- project_active_squad fetches the assigned manager owned item and passes
it to the projector (non-fatal on error/absence).
- handle_put_squad authorizes the resolved manager against the active
club (like a slot) and persists it via set_squad_manager after the
atomic squad replace; fails loudly, never silently drops it.
Move the squad manager from an opaque, unvalidated wire ref in the squad
extension to a resolved, ownership-backed assignment (Core migration 0023
squad_managers).
- squad::to_proposed reverse-resolves the manager wire ref to a Core
owned_card_id on ProposedSquad; an unresolvable manager is reported as
an unresolved wire id (a live save is refused rather than assigning a
manager the club does not own).
- squad_ext: drop the opaque manager field from Fifa17SquadExtensionV1
(clean cutover) and the now-unused SquadEntityRef->WireItemRef From.
- squad_projection: project the manager as the STATIC_REVERSED [{id,dream}]
wire ref resolved from the owned assignment; absent -> [] (never faked).
Richer manager itemData (contract/league/nation) is INFERRED-only and
left out pending wire reversal.
- import(apply): drop a historical dangling manager ref rather than
failing the whole import (live PUTs still refuse an unresolved manager).
AGENTS.md is the new canonical AI-agent entry point (FIFA17 active target,
repo map, FIFA23->FIFA17 pivot history). README/CLAUDE/setup.sh get stale
banners pointing to it. Retained pre-existing WIP brought forward.
Reads openfut-core's authoritative API directly (balance, entitlements, profile,
club, collection, squad/ext) with the fifa17 game header and emits a
machine-readable JSON snapshot plus integrity checks: coin cross-check, duplicate
owned_card_id, squad-references-non-owned, owned-set + card-multiset hashes for
drift/resurrection detection across operations and restarts. Read-only oracle
tooling; used to verify Phases 1-9 of the Core-backed FUT loop audit.
openfut-hook builds two mutually exclusive injection paths from one crate. The
default (FIFA 23) path installs getaddrinfo/connect/ProtoSSL/origin_spy transport
hooks; `--features fifa17` installs only the FIFA-17-safe logic (module map,
FIFA 17 cert-verify, SBC dispatch, store tab bind).
Deploying a default-feature build into FIFA 17 hijacks the login transport: the
client reports "Unable to connect to the EA servers at this time" and none of the
FIFA 17 repairs are present in the binary at all.
That happened today: artifact 1c71a17a was built by hand without the feature and
deployed, costing two failed launches. It was diagnosed only by comparing embedded
strings between the deployed DLL and the last known-good one (the deployed DLL had
0 occurrences of CardsDLL_Win64_retail.dll and SBC_DISPATCH, and 6 of cert-verify
plus 1 of "connect: inline-hooked" -- the inverse of a fifa17 build).
`build` already passes --features fifa17, but OPENFUT_FIFA17_HOOK_DLL lets a
hand-built DLL reach stage/deploy, so verify_fifa17_profile asserts the profile on
the bytes: CardsDLL_Win64_retail.dll and SBC_DISPATCH must be present, and the
FIFA-23-only markers must be absent. Wired into verify_inputs (stage/inspect) and
into deploy's staged-artifact checks.
Verified: the gate rejects 1c71a17a, accepts 3641d581 (last known good) and
f0ef528f (the corrected fifa17 build now deployed).
Records the store-entry category-clamp hook (launcher 9aecc65) as the
launcher tip on main. Only the launcher gitlink is bumped; core stays
271c363 and bridge stays 0f581eb (both as already recorded), and the
in-tree formatting/doc churn is left untouched.
Reversed from the pinned CardsDLL (4706a881): eligibilityKey and
eligibilityOperation are localization ordinals (LOC_SBC_ELG_KEY_%d),
not the atom hex ids. The client's only consumer is the requirement-
display string builder at ~0x1800ef900 (formats via indexed locale
keys, no comparison/gate). The ordinal->string map lives only in the
packed locale (absent from all assets we hold), so any emitted value
would render the WRONG requirement text. Submission stays fully
validated server-side by Core; the empty elgReq is display-only.
Correct ENDPOINT_MAP.md's implied atom-id==ordinal assumption and
pin the exact remaining blocker at the emit site.
Picks up openfut-launcher 94feaec, which makes the guarded native SBC dispatch
repair a build-armed promoted feature instead of an OPENFUT_SBC_DISPATCH env gate.
Any launch path (Steam, the launcher Launch button, bare umu-run) now gets the
repair, so the SBC screen no longer depends on a harness script exporting a
variable. Every runtime guard is unchanged; rollback is a version.dll file swap.
A reward pack advertised its own id (70-75) as assetId, which is not a client art
asset, so its My Packs tile rendered blank. Reward tiles now carry their tier
store pack as the cover asset (bronze->1, silver->3, gold->5) while `id` stays the
pack own id (the open packId / SERVER_ID); the sentinel keeps its own id so it
stays an inert placeholder.
LIVE-MAPPED on the retail client 2026-08-18 across all three store tabs and two
reward tiles, which corrected an earlier wrong assumption:
* assetId only gates whether art renders AT ALL (unknown id -> blank tile).
* WHICH art is drawn comes from packType + packContentInfo.rareQuantity, NOT
assetId: BRONZE+1rare -> bronze card, BRONZE+3 -> silver, SILVER+1 -> gold,
SILVER+3 -> silver trio, GOLD+1 -> blue special, GOLD+3 -> red inform.
Remapping assetId 5->3 and 3->2 left both frames unchanged and only rotated
the featured player, which proves art is content-driven.
So a reward tile now shows the same cover as the equivalent purchasable pack (a
gold reward shows the blue-special art the 5000-coin Gold Pack shows - EA art
advertises an aspirational card rather than the tier colour). Regression test
reward_tiles_carry_a_renderable_cover_asset added; pack70 golden regenerated.
Reward/My-Packs tiles were emitted with no currencies row (dropped to avoid an
"undefined" payment label). But FIFA 17 opens My Packs through the store PURCHASE
flow (My Packs is a client-side filter over the store catalogue; the open-vs-buy
fork is client-side and no response selects it), so a tile with no purchase path
is not actionable — clicking it navigates instead of opening, sending no request.
Fix: keep the coins currency at the pack price (0 for reward packs; no extPrice,
so no `or %1s` mtx bug). The client then treats the tile as free and clicking
sends POST /purchased, which the server opens for free (owned-only redeem, no
debit). The empty-My-Packs sentinel keeps no currency row so it stays
non-openable. pack70 golden regenerated.
LIVE-PROVEN 2026-08-18: a reward Silver Pack opened and revealed cards on the
retail client (host log: POST /purchased/items 200 -> GET /purchased/items).
SBC completion and the other Core reward services grant packs with symbolic
definition_ids ("silver_pack", "gold_pack", ...). The FIFA 17 pack system keys
entirely on numeric catalogue ids, and entitlement_pack_ids / handle_pack_open
resolved definition_ids with definition_id.parse::<u64>(), so every symbolic
reward pack was silently dropped from the openable My Packs list. The unopened
count (recoveredPacks, = entitlement count) still counted them, so the client
showed "you have N packs" but had no tile to open -> "no pack available".
Fix (adapter-layer, Core stays game-neutral): add owned-only reward pack
catalogue entries 71-75 (bronze/silver/gold/rare_gold/icon) and a resolver
owned_pack_id_for_definition() that maps both numeric owned ids and the symbolic
reward names to their numeric owned-only pack. entitlement_pack_ids and the
pack-open entitlement selection now use it, so reward packs render as openable
My Packs tiles and redeem their entitlement for free (consume-once, no debit).
Server-verified on staging: 3 Core reward entitlements now render 3 openable
mypacks tiles matching recoveredPacks=3. Tests: adapter resolver + rendering,
host symbolic-reward open flow; full adapter + host suites green.
The FIFA 17 client gates challenge re-entry on timesCompleted, not on the
repeatable flag: a nonzero count renders the tile COMPLETED and refuses re-entry
even when repeatable=true. So a repeatable challenge now always projects
timesCompleted=0 (challenges_body) and its set as challengesCompletedCount=0
(sets_body); a non-repeatable challenge keeps its true count and stays locked
once completed. Core keeps the authoritative completion record — economy is
unaffected; this is presentation only.
Live-proven on the retail client 2026-08-18: a completed repeatable Bronze
Upgrade now re-opens for a fresh submit instead of blocking. Regression test
repeatable_completed_challenge_stays_enterable added; adapter + full host suite
(incl. differential and the concurrency race) green.
The all-groups landing (shown on first store entry) renders one tile per group
whose background is the client-bundled packs_backgrounds_%d.dds, selected by
displayGroupAssetId. Live-probed on the retail client 2026-08-18: index 0 is
blank; indices 1/2/3 render real pack art. Assign non-zero per category
(bronze=1, silver=2, gold/mypacks=3) so that landing shows native pack art
instead of blank shields. The persistent tabbed store (MY PACKS/BRONZE/SILVER/
GOLD PACKS) draws pack art from the packs themselves and is unaffected.
LIVE-CONFIRMED end-to-end: native tabbed store renders the real 6-pack catalogue
with correct prices (Gold 5000 / Premium Gold 7500), counts (12 items, 10 gold,
1/3 rares), pack art, and no "or %1s" line.
- store_catalog: replace the invented catalogue with the real always-available
FUT17 regular packs (Bronze/Prem Bronze/Silver/Prem Silver/Gold/Prem Gold) at
real prices + tier composition; PackDef now carries per-tier quantities.
- pack_body: drop extPrice (its mtx side-effect switched on the broken "or %1s"
FIFA-Points tile line; plan-2026-08-05-store-subsystem.md section 3.4).
- pack_content: tier-aware generator draws each pack bronze/silver/gold
composition with special_chance bias + empty-tier fallback.
- host: CoreAccess::all_definitions (GET /cards); build_content_pool draws the
FULL card universe via non-minting catalog lookup, owned-inventory fallback.
- economy_differential: store ops reclassified DIFFERENT-BY-DESIGN (Rust is the
authoritative store; Python oracle stays the untouched rollback baseline).
- fixtures/tests updated to the real catalogue.
Odds are DESIGNED placeholders (FUT17 pack probabilities were never published);
club items remain excluded (cardtype-9 mapping unknown). Full regression green;
real prices + tier-correct draws verified server-side on staging.
Retail Gate C captured PUT /sbs/challenge/101/squad: 23-slot players[] of
{index,itemData:{id,dream}} plus manager/chemistry/rating/formation siblings.
parse_wire_item_ids already handles it (players[].itemData.id, non-zero only);
update the stale "captures unavailable" note and add a verbatim regression test.
Runs the REAL roster_server.py under `sudo unshare -n` (so port 8081 is free and
production is never touched) and validates its certificate BY THE DIALED IP, proving the
before/after the SAN fix (fbc0da2) targets:
* OLD cert (DNS-only, production's current shape) -> a by-IP-verifying client is
rejected with "IP address mismatch, certificate is not valid for '127.0.0.1'" — the
certificate_unknown class the FIFA client hit.
* NEW cert (fixed generator, DNS + IP SANs) -> verifies through the actual roster
server and returns the roster XML (200, application/xml).
roster-cert-verify.py is the client probe (trusts the served self-signed cert as CA,
checks it against the dialed IP, then GETs /fifa17/fut/rosterupdate.xml).
roster-cert-iso-test.sh drives the real server with each cert and asserts new=pass,
old=fail. Two harness bugs were found and fixed while writing it (a shared /tmp log the
production run owns, and a subshell pid that left the first server alive so the "old"
probe hit a stale server presenting the new cert — the tell was "self-signed" instead
of "IP mismatch"), so the final before/after is clean.
Complements the in-process check: this exercises the production server code path, not a
hand-rolled server. Live production confirmation still needs the container rebuilt with
OPENFUT_ADVERTISE set (operator-gated).
The entire docs/ tree (24 top-level notes, 68 evidence captures, 2 plans, research) has
been migrated into ~/OpenFUT-Vault, the curated Obsidian vault, which is now the sole
home for project documentation. Merges preserved all detail (obsolete material kept
under "Superseded" sections); evidence/plans were copied byte-identical; every migrated
note records its Source: docs/<original>.md provenance. Vault commit f55a5ba.
Documentation lives in the vault from here on. Code/script comments that still reference
docs/ paths are stale pointers only (no build dependency); they can be repointed at the
vault opportunistically. References to fifa17-recon/docs/ are a different tree and are
unaffected.
Updates status from root-caused to fixed, and records that option 1 (IP SAN) was
taken across the three cert generators, with the verification and the operator-gated
production rebuild that remains.
The FUT hub failed to load with "An error occurred downloading the FUT Squad
Update" because the client dials the roster (https://<advertise>:8081) and the
redirector BY IP, while the served certificate carried DNS SANs only
(winter15.gosredirector.ea.com + wildcards). The client aborts that handshake with
fatal certificate_unknown. Root cause and evidence in
docs/FIFA17_FUT_SQUAD_UPDATE_TLS.md (commit 082246c): a wire capture shows the client
offering TLS1.2 with RSA suites, the server selecting them, then rejecting the cert —
and autopatch demonstrably patched both ProtoSSL gates in that process, so this
validation path is NOT one of the two the client-side patch covers. The SAN is the fix.
Three generators produced the cert and none put the advertised IP in the SAN:
* docker entrypoint.sh — the production path. The advertised IP is a RUNTIME value
(OPENFUT_ADVERTISE), unknown at image-build time, so the cert is now reconciled at
startup: reissued with IP:$ADV,IP:127.0.0.1 in the SAN only when the current cert
lacks it. That makes a restart reuse the same cert (no per-start fingerprint churn,
which would otherwise recreate the Aug-13 surprise) and self-heal if $ADV changes.
* Dockerfile — installs openssl unconditionally so the entrypoint can reissue at
runtime (previously it was dropped with the apt lists), and bakes a loopback-IP
baseline cert so a plain `docker build` still yields a usable image.
* openfut-fut.sh — the local orchestrator. ensure_cert now defaults the SAN IP to this
host's primary LAN IP (OPENFUT_ADVERTISE overrides) and reissues when the cert lacks
it, instead of only generating when the file is absent.
Verified without the client, which is the strongest evidence obtainable here: a
verifying TLS client checking the cert BY IP rejects the old DNS-only cert ("IP address
mismatch, certificate is not valid for '10.10.0.120'") and accepts the new
IP-bearing cert; and the entrypoint reconcile is idempotent end to end — an old cert is
reissued to carry IP:$ADV, a simulated restart leaves the fingerprint unchanged, and
the final SAN carries both the advertised and loopback IPs.
Live confirmation needs the production container rebuilt with OPENFUT_ADVERTISE set
(operator-gated); production is otherwise untouched.
entrypoint.sh carries unrelated pre-existing uncommitted work (env-based component
selection) that is not on any branch; only the cert-reconcile block is committed here,
and that work is left intact in the working tree.
The launcher spawned `python3 lsx_responder_v2.py` and `python3 autopatch.py`. Both are
now Rust workspace crates, and the launcher spawns the binaries (gitlink 1cd4f18).
openfut-lsx (2244 lines, 57 tests) — EA Origin LSX emulator on loopback 4216.
Dependency-light on purpose: `aes` for the one security-shaped primitive, parking_lot
per the project lock rule. AES-128-ECB is the whole cipher requirement, so the
surrounding framing (PKCS7, lowercase hex, NUL-termination) stays explicit and separate
because it is protocol, not cryptography.
openfut-autopatch (43 tests) — ProtoSSL cert gates plus the CardsDLL store patches,
applied over /proc/<pid>/mem. Deliberately dependency-free: a tool that writes another
process's memory should be auditable end to end without a dependency tree. std has no
getuid and no local-time formatting, so it carries a small TZif reader rather than
pulling in chrono to reproduce Python's strftime('%H:%M:%S').
The Python remains in fifa17-recon/tools. It is NOT dead: the docker entrypoint,
client_arm.sh, the runbooks and test_autopatch_guard.py still use it. Only the
launcher's dependency on Python is gone, which is what was asked for; deleting the
recon toolchain's implementation would have broken unrelated workflows.
VERIFICATION — the ports are checked against the Python, not against themselves:
* Crypto parity across THREE implementations. The Rust tests assert the Rust's own
constants, which proves consistency, not parity, and the Python cannot run here
(pycryptodome absent) with the client host unreachable. So the LCG and key derivation
were transcribed from the Python and run as plain arithmetic, and every AES value came
from the openssl CLI. All agree: msvcr_rand(7)==61, _TAIL_CONST
954f64f2e4e86e9eee82d20216684899, the 96-hex emu challenge shape, the derived session
key 6a9da3e78615153cc2f10eec25ae6382, the framing rule at both boundaries (an aligned
payload gains a whole block), and the port's pinned 4-block login-frame ciphertext.
* LSX end to end on the real port. 4216 here is a docker forward into the production
netns, so the smoke test runs under `unshare -n` — the real binary on the port the
client actually dials, with no port-override hack and no risk to production. A
hand-written client read the unprompted <Challenge>, completed the handshake, and
decrypted the GetProfileResponse (PersonaId 33068179, Persona CAGE) with a session key
derived INDEPENDENTLY of the Rust, then observed the Login pushes across all three
candidate senders.
* autopatch behaviourally. The startup banner, the --launcher-pid watchdog exiting with
the exact Python message, dual stdout+logfile output, and a missing value rejected
with Python's own "invalid --launcher-pid". The subagent additionally cross-checked
every constant by executing the Python module and drove the binary against a synthetic
client (correct comm, a CardsDLL mapping, gates mmapped at their absolute VAs),
confirming all eleven patches byte-exact in table order.
* The `[store-guard] verified capability …` line is byte-identical to openfut-launcher's
own parser fixture, so backend capability registration still works.
Workspace builds; openfut-lsx 57, openfut-autopatch 43, openfut-launcher 74 tests green.
Recovered the client's own dialog text from memory rather than inferring from the
server, which is what finally identified the subsystem: "An error occurred downloading
the FUT Squad Update" is the ROSTER update, not the player's lineup. Four squad-shaped
fixes before that were aimed at the wrong thing.
Wire capture shows the client aborting the handshake itself: it offers TLS1.2 with RSA
suites, the server selects TLS1.2 and sends its certificate, and the client replies
fatal certificate_unknown. So protocol and ciphers are compatible and the certificate
is the problem. That certificate is DNS-SAN-only while the advertised ROSTERUPDATE_URL
is an IP literal, and it was regenerated Aug 13 -- after the Aug 12 session being used
as the known-good control, which therefore says nothing about the current cert.
Notably autopatch DID patch both ProtoSSL gates in the failing process (log line plus
live bytes reading back patched) and the client still rejected, so those gates do not
govern this path -- contradicting roster_server.py's standing comment that they make
self-signed certs acceptable.
Documents what was ruled out with evidence (hub route shapes, squad shape, squad
round-trip, advertised hosts, TLS version, Blaze health), the probing gotcha that a
default modern TLS context misreports this server as broken, the unresolved question of
why production appears unaffected, and three fix options with a recommendation. No fix
applied.
Three diagnostics from chasing a FUT error that four server-side fixes failed to
resolve, kept because the technique generalises.
client-error-string.py recovers FIFA's on-screen message from /proc/<pid>/mem,
read-only, scanning ASCII and UTF-16LE (FIFA UI strings are wide). This ended the
guessing: the dialog reads "An error occurred downloading the FUT Squad Update.
Please try again." -- a CONTENT DOWNLOAD failure, not the player's lineup. Every
squad fix before it was aimed at the wrong subsystem, because "squad update" in FIFA
means the roster update, and the server-side symptom (a squad the client would not
accept) was consistent with both readings. When the server says 200 and the client
says no, the client's own words are the cheapest evidence available and should have
been the FIRST thing recovered, not the fifth.
hub-dump.py + hub-diff-prod-staging.py diff every hub route between production
(known-good, same client accepts it) and staging, comparing key presence and JSON
types rather than values, since values legitimately differ. Result: 0 structural
differences across 14 routes, which retired the whole "a missing field breaks
bootstrap" line of investigation in one run instead of one restart at a time.
Also ruled out with evidence: cert gates ARE patched (autopatch logs
"pid 56298: PATCHED cert gates", and the gate bytes read back as the patched
patterns); the roster server serves the FUT Squad Update fine (TLS1.2
AES256-GCM-SHA384, HTTP/1.0 200, application/xml) once probed with
ALL:@SECLEVEL=0 -- a default modern context gets SSLV3_ALERT_HANDSHAKE_FAILURE and
would have been a false alarm; production and staging Blaze advertise identical
roster/POW hosts; the Blaze session is healthy and answering PINGs; and the squad
round-trips exactly through PUT/GET.
Adds the manager reference, the last remaining difference from the squad the same
client demonstrably accepts. Staging's squad shape is now identical to production's:
zero missing keys, zero type differences, zero empty-vs-populated mismatches.
The manager looked unfixable. Production points at instance 100000427 while the
staging club holds 11 players and zero staff, so there was apparently nothing to
reference, and inventing an id would have pointed at a non-existent item.
Checking production properly dissolved the problem: 100000427 is absent from
production's OWN club listing too. /club/staff returns 1975 items spanning ids
100000001..100004826 and 100000427 is not among them, and the type=staff/type=manager
filters are ignored (200 players either way). Production's manager reference is
dangling and the client accepts that squad anyway, which proves the client does not
validate the manager id against the club -- only a populated array matters.
So the reference is mirrored verbatim, dangling id included. That replicates the
known-good state exactly and is better than pointing the manager slot at a player,
which would have been a guess dressed up as a fix.
Method note: every step here came from diffing against production rather than reading
the client. The host reported squad-active 200 outcome=ok throughout, and 200 with the
right players was never evidence the client accepted the body.
First seed sent only squadName/formation/captain/players. The host logged squad-active
200 outcome=ok and /squad/0 showed 11 occupied slots, yet the client still threw a FUT
squad update error -- a 200 with the right players is not proof the client accepts the
body.
Diffed against production's known-good squad, which the same client accepts, comparing
key presence and JSON TYPES rather than values. Staging returned null for exactly the
five fields the PUT never carried, because the extension stored nothing for them:
squadType (a string enum), chemistry, rating, starRating (ints) and custom (the opaque
33-int tactics array the client definitely parses). kicktakers was empty where
production carries five.
Now sends all of them: squadType REGULAR_SQUAD, chemistry, rating/starRating derived
from the XI's mean rating, production's custom array verbatim (opaque server-side, only
its shape matters), and five kicktakers. Re-diff leaves exactly one difference --
manager, which production points at owned staff instance 100000427 while the staging
club holds 11 players and zero staff. Left empty rather than inventing an id that
references a non-existent item; recorded in the code as the one known remaining gap.
Also fixes a KeyError from the rewrite dropping the players key.
The sold A/B stalled twice on preconditions no headless check exercised: the staging
identity had no squad (the hub refuses to open, showing a squad update error), and any
route without a Rust owner falls through to a deliberately dead Python upstream and
answers 502. Each cost a full operator cycle.
Sweeps the routes the client is observed to request and separates three failure classes
that need different fixes: 502/PYTHON_FALLBACK (no Rust owner), missing_integrity (200
but the underlying state is absent -- exactly 'no extension stored' before the squad
was seeded), and 200-but-unusable (a squad with zero occupied slots). A 200 is not
proof the client is satisfied, so squad responses are judged on occupied slots.
Also reads the host's own classification for the requests just made, since the host is
the authority on ownership and integrity rather than the response body.
Every path is verified against what the client actually sends. A first pass flagged
five 'fatal' routes that were my own guesses -- /accountinfo (client uses
/user/accountinfo), bare /squad (uses /squad/active), and /watchlist (camelCase
watchList). Crying wolf about the stack is worse than not checking, so the list now
carries only observed paths and that trap is written down in the comment.
Current result: 14 ok, 0 integrity warnings, 0 fatal.
The A/B identity had owned items but no squad, because the sold-row work only ever
needed the tradePile wire. Every headless check passed -- none of them asks for a
squad -- but the real client refuses to enter the FUT hub with an empty one and shows
a squad update error. A squad is a hub precondition, not a Transfer-List detail.
Seeds via the real PUT /ut/game/fifa17/squad/0, the same request the client sends, so
parse_squad_put/build_squad_write produce exactly what a genuine save would. Writing
Core rows by hand could yield a shape the live path never emits, which is the kind of
divergence that quietly invalidates an experiment.
Picks one owned player per 4-3-3 slot, best rating first, without reusing an instance;
fills the fixed 23-slot array with 0..=10 as the pitch and empty slots as
itemData.id == 0; refuses to write a partial XI and reports any out-of-position
substitution loudly rather than silently reproducing the broken state. Verified
0 -> 11 occupied with no substitutions and a {"id":0} ack.
Bridges openfut-launcher c542415 onto the already-built launcher on .105 by writing
WINEDLLOVERRIDES=version=n,b into game_profile.env, which that build does apply.
Forward-compatible: the fixed launcher defers to a profile that already pins
version=, so this value simply wins.
Records the prior value -- including its absence, as the literal <absent> -- to a
sidecar before mutating, so revert restores the real previous state instead of
assuming the key was missing. Refuses to edit while the launcher runs, since it holds
its config in memory and would write the stale value back.
Without version=n,b the launcher's own launch path never loaded the version.dll hook
proxy, so the Blaze ports it writes to openfut.cfg were ignored and the client
silently reached production via /etc/hosts instead of the configured server.
openfut.cfg is not the source of truth for the client's Blaze ports -- the launcher
is. It reconciles openfut.cfg from ~/.config/openfut-launcher/config.json,
fail-closed, immediately before every launch. So `staging` set the ports, verified
them, and the next launch silently reverted them.
Caught only because the capture harness cross-checks instead of trusting the screen.
The operator reported "the Transfers tile does not show Sold" -- which looked like a
clean negative result about the sold counter, and was in fact a reading of their
PRODUCTION club, where sold:0 is correct. Evidence chain:
* route-log delta contained 5 lines, all of them the harness's own GETs; the
client issued nothing to staging at all;
* `ss -tnp` on the client showed FIFA17.exe pid 39482 ESTAB to 10.10.0.120:42130
(production Blaze) plus TIME-WAIT to :8099 (production UTAS);
* the hook logged `blaze_redir=42127 blaze_main=42130`;
* openfut.cfg mtime was 2s before process start, sha back to the production value.
Had the harness reported the tile at face value, the sold counter recovered from
CardsDLL would now be recorded as refuted by a run that never reached the code.
Fixes: own the launcher config (source) before openfut.cfg (derived), with the same
record-before-mutate sidecar discipline on both; refuse to edit while the launcher is
running, since it holds config in memory and would write the stale values back;
report both files and both guards in `show`. launcher_running() matches the
kernel-truncated comm "openfut-launche" -- the full name exceeds 15 chars, which has
bitten this project before.
No production change; staging stack and its variant-A sold row untouched.
`pgrep -f FIFA17.exe` matched the remote shell executing it -- the SSH command line
contains the literal pattern -- so fifa_running() always returned True and the port
switcher could never edit openfut.cfg. It refused with "REFUSING to edit ... while a
FIFA client is running" moments after FIFA had actually exited.
Fail-closed, so nothing unsafe happened, but the guard was permanently stuck and
blocked the A/B entirely.
Now matches /proc/<pid>/comm exactly, which is the executable name: the invoking
shell reads as zsh and cannot self-match, while a genuine FIFA process still does.
Validated both directions with the same loop -- it found pid 36958 while FIFA was up,
and reports gone once it exited. Still fail-closed on read errors.
The lesson generalises: a pattern-matching process guard checked over a transport
that carries the pattern in its own argv is self-satisfying, and a guard that can
only ever say "yes" is not a guard.
Turns the operator's job into "navigate, say go" and removes any chance of a
half-recorded variant. One command captures and labels: the staging wire surfaces,
the client's OWN auction record decoded read-only from /proc/<pid>/mem (STATE,
YOURBID, COINS_AWARDED, MIN_CREDITS, IS_GLOW, INBOX, CARD_OFFERSTATE), and the
staging host route-log DELTA since the last capture -- which is how a client-issued
DELETE .../trade/sold gets OBSERVED rather than assumed.
The part that matters is the wire-vs-memory cross-check. It validates the
observation mechanism against a known-positive in the SAME run: if the wire says
bidState "highest" and the client's memory decodes 2(highest), the probe is
demonstrably reading the right struct this time. It also recomputes the native
IS_GLOW/INBOX formulas from the wire and compares them to what the client stored.
Proven honest on first run: with the client attached to PRODUCTION and not on the
Transfer List, it reported the staging sold row on the wire, 0 client records, and
INSTRUMENTATION NOT VALIDATED -- refusing to draw a conclusion from an empty read.
Two earlier sessions were misled by exactly that (a sampler bug printing
"countdown NO", and auction containers read while the screen was unbound), so an
empty container is explicitly not treated as an empty pile.
Probe base-address discovery was separately confirmed against the live client
(pid 36958, FNV control=MATCH, model resolved, containers read cleanly), and
production's wire independently agreed at total=0.
No production change. Client config untouched (still production Blaze ports).
The brief's gate: if the harness varies bidState AND coinsProcessed together, the
client's reaction is attributable to neither. The env knobs were already orthogonal
(--variant and --coins-processed are independent, cp defaults to 0), but
sold-wire-check.py was flipping BOTH for variant B as a convenience, which is exactly
the contaminated A/B the brief forbids. Fixed: the primary pair now holds
coinsProcessed at 0 and asserts the differing-field set is exactly ['bidState'].
New scripts/sold-ab-differential.py is the pre-live gate. It settles ONE synthetic
sale, then re-reads every seller-facing surface under each variant by restarting only
the host (same Core, same DBs, same sale), and diffs with explicit classification --
MISSING / EXTRA / TYPE_MISMATCH / VALUE_MISMATCH -- rather than a boolean "equal?".
Two orthogonal pairs:
PRIMARY bidState highest vs buyNow, coinsProcessed held at 0
ORTHOGONAL coinsProcessed 0 vs 1, bidState held at highest
Result, 36/36: the ONLY finding on /tradePile is
VALUE_MISMATCH auctionInfo[0].bidState A='highest' B='buyNow'; /trade/status differs
in exactly the same one path; counts are byte-identical. The orthogonal pair's only
finding is auctionInfo[0].coinsProcessed. C_cp0's sha256 equals A_highest's, so the
capture is reproducible rather than merely consistent.
Counts states the live run has to interpret, measured not guessed:
S1 0 active + 1 sold -> count 0, selling 0, sold 1
S2 1 active + 1 sold -> count 1 (active mode) vs 2 (membership mode)
That divergence IS the open question for the client; production is unchanged.
scripts/sold-client-ports.py switches ONLY the two client Blaze port lines, and is
built so restoration cannot depend on memory: it records the production values to a
sidecar on the client BEFORE the first edit and restore reads that sidecar, refusing
if it is absent. It rewrites only known keys (a missing key is an error, never a
silent append), re-reads and verifies afterwards, and REFUSES to edit while a FIFA
client is running because the hook reads the file at connect time.
Phase 0 evidence under docs/evidence/sold-ab-2026-08-18/ with a sha256 per surface,
one file per variant so A can never overwrite B.
Live client A/B NOT run: a production FIFA session is currently live on 10.10.0.105
(pid 32188), and live-session mutual exclusion applies. The client config was NOT
touched -- the switcher's guard refused, as designed.
Production untouched: prod-host pid 3631953, coins 29,843,976, /tradePile 0,
counts.sold 0, club 1966; nothing under /home/alex/openfut-promotion/state/ opened.
Static RE exhausted CardsDLL on the one open question: for a closed row
IS_GLOW = (bidState != none) and INBOX = (bidState in {highest, buyNow}), so
closed/highest and closed/buyNow are BIT-IDENTICAL natively. But bidState is
published to the movie verbatim as YOURBID, so the FUT ActionScript CAN separate
them. This builds the controlled experiment that asks the client which one it
treats as the seller's sale.
PRODUCTION SAFETY IS THE FIRST CONCERN
New module openfut-utas-host/src/sold_experiment.rs. Every knob is OFF unless its
env var is set, an unrecognised value is OFF rather than a default token (silently
picking one would fabricate the answer being measured), and the host logs a startup
banner naming the active variant so a staging capture can never be mistaken for a
production one. With no env set, /tradePile and /trade/status emit only real active
auctions (the Fix A invariant) and counts still report sold: 0. The entire existing
test suite now passes SoldExperiment::OFF explicitly, making it a regression guard.
OPENFUT_FIFA17_SOLD_EXPERIMENT = highest | buyNow (else OFF)
OPENFUT_FIFA17_SOLD_COINS_PROCESSED = 1 (else 0)
OPENFUT_FIFA17_SOLD_COUNT_MODE = active_plus_sold (else active)
WHAT THE EXPERIMENT PROJECTS
Uncleared sold listings appear in /tradePile and /trade/status as tradeState
"closed" with the token under test and currentBid = the sale price; counts report
the real sold tally. There is ONE record builder, so the A/B changes only what is
passed into it, and a test asserts that EXACTLY ONE field differs between the two
variants -- without that control the client's reaction is not attributable to the
token and the whole experiment is void. coinsProcessed (Flash COINS_AWARDED) varies
independently so the third pass cannot be confounded with the first.
CLEAR-SOLD, PE-PROVEN
New EconomyRoute::MarketClearSold for DELETE .../trade/sold, classified BEFORE the
generic trade cancel arm -- a `sold` tail carries no id, so the cancel handler would
have parsed nothing and acked while clearing nothing. Builder 0x1801647c0 emits
"/sold" when the tradeId field is zero and "/%lld" otherwise; the client calls it
RemoveAllSoldFromTradePile. New market-store column cleared_at records the seller's
acknowledgement SEPARATELY from the sale, so clearing can never be mistaken for
re-settling: it is presentation only, moves no coins and no ownership, and is
idempotent for client retries.
FOUND AND FIXED A LATENT STORE BUG
Adding a column via the additive ALTER path immediately after CREATE TABLE in the
same open() desynced sqlx's per-connection schema cache: a fresh store then read a
12-column row while metadata said 13, panicking a pool worker with an index
out-of-bounds and silently returning zero listings. Declaring cleared_at in
CREATE_LISTINGS fixes it; the ALTER now only serves pre-existing stores. This would
have bitten the next column too.
STAGING, WITHOUT TOUCHING PRODUCTION
The client learns the UTAS base from BLAZE (blaze_responder_v3b.py:646 hardcodes
:8099), and it dials that port directly, so redirecting UTAS means changing Blaze or
port 8099 -- both production. 10.10.0.121 is unreachable. The compliant path is a
parallel stack on spare ports plus a one-line change to the CLIENT's own config:
* scripts/sold-staging-up.py / sold-staging-down.py -- staging Core 18081,
utas-host 8299, Blaze 42327/42330/42331 advertising :8299, two seeded identities,
own DBs under /home/alex/openfut-sold-staging/. Patches a COPY of the Blaze
responder and asserts every substitution applied, so a silent no-op cannot leave
it pointing at production. Kills only recorded pids whose cmdline contains the
staging dir (openfut-utas-host matches BOTH, so pkill-by-pattern is banned).
* docs/SOLD_STAGING_RUNBOOK.md -- the exact client change and its revert.
* src/bin/staging_sell.rs -- the synthetic Buyer B, running the REAL settlement
(CoreEconomy::settle_sale) then mark_sold. Settle-first ordering: a failure
leaves the listing live with nothing moved. Refuses any path containing
openfut-promotion or the production ports.
* scripts/sold-wire-check.py -- proves the whole flow headless before any operator
time is spent.
WIRE CHECK: 35/35 PASS on the canonical 150-coin sale. Seller 1,000 -> 1,143 (fee 7,
proceeds 143), buyer 20,000 -> 19,850, ownership transferred, exactly ONE
authoritative instance, economy shrank by exactly the fee. Sold row: closed,
currentBid 150, expires 0, twelve atoms, counts sold 1 / selling 0, /trade/status
agreeing. Variant B differs only in bidState and coinsProcessed. Clear: 200 {}, row
gone, counts.sold 0, no coins moved, buyer keeps the item, second clear a safe no-op.
Gates: 104 host lib tests (+9), all 7 host targets green, clippy clean, zero fmt
diffs in the new code. Settlement candidate unchanged. NOT PROMOTED.
Production untouched: prod-host pid 3631953 uptime 2h44m restarts=0, coins and
/tradePile unchanged, nothing under /home/alex/openfut-promotion/state/ opened.
The A/B itself is NOT yet run: it needs a real FIFA client, which is operator work.
Task A, static phase. Ghidra 12.1.2 headless via the repo's own pyghidra harness
over CardsDLL_Win64_retail.dll (13,382 functions). Queries and raw decompiler
output committed under docs/evidence/market-sold-re-2026-08-17/.
RECOVERED FROM THE BINARY
1. No sold token, now EXHAUSTIVELY: both vocabularies dumped to their sentinels
rather than sampled. tradeState is exactly 4 rows; itemState is exactly 12
(invalid/free/WAITING_FOR_GAME/inGame/forSale/offered/activeBadge/
activeHomeKit/activeAwayKit/activeBall/activeStadium/active=255). A sold row
MUST therefore be a combination of existing atoms.
2. What closed does, complete, from the auctionInfo deserializer 0x18013e410:
IS_GLOW = (tradeState==closed) ? bidState != none
: bidState in {outbid, buyNow}
INBOX = bidState in {highest, buyNow}
3. The full record -> Flash map from the publisher 0x1801bf030, superseding the
partial list. The prize: record +0xbf is published as COINS_AWARDED, fed by the
coinsProcessed atom 0x2f4. The corpus had recorded that atom's type and noted
its consumer was never found; it is now traced. DURATION also renders the
localised FUT_AUCTION_EXPIRED when expires underflows.
4. highest vs buyNow on a closed row is UNDECIDABLE from CardsDLL, by proof: both
yield IS_GLOW=1/INBOX=1, bit-identical. But bidState is ALSO published verbatim
as YOURBID alongside STATE and COINS_AWARDED, so the movie does receive the raw
values - the discrimination exists and lives entirely in unread ActionScript.
This retires the question as a static target, and it contradicts the
third-party lore that a seller's sold row is closed+buyNow (the corpus's own
lifecycle table says closed+highest and assigns buyNow to the buyer).
5. The clear-sold verb EXISTS. Builder 0x1801647c0 emits "/sold" when the tradeId
field is zero and "/%lld" otherwise, on route base ut/delete/%s/trade, response
class RS4 FutISRemoveTradeServerResponse. Confirmed by the client's own
request-name table entry RemoveAllSoldFromTradePile. A BULK clear-sold verb only
makes sense if sold rows PERSIST in the seller's pile until cleared, which is
incompatible with our Fix A invariant - so the sold path will require revisiting
it under live validation.
6. The seller's SOLD counter is real, proven end to end with no inference: the hub
tradePile sub-deserializer 0x18013ead0 writes atom sold 0x2c9 to +0x1d8, and the
tile publisher 0x1800b1dc0 renders +0x1d8 as Flash TEXT3 under the localised
caption FUT_TF_SOLD. Siblings: selling -> +0x1d2 -> FUT_TF_SELLING,
count -> +0x1d4 -> FUT_UC_ITEMS, plus FUT_TF_WINNING/FUT_TF_OUTBID on the
Transfer Targets tile. We and the Python oracle both hardcode sold:0, so that
bucket can never fill.
7. Reusable method: an atom id is the INDEX into the alphabetical atom-name pointer
table at base 0x1802d2760. Validated 12/12 against the known auctionInfo atoms
and cross-checked against fifa17-recon/docs/fut_atoms.tsv. Documented gotcha:
resolve a name by the pointer slot INSIDE the table, never by the first matching
string in the binary, or you get confident nonsense.
8. An auction-outcome vocabulary exists (auctionSoldBid 0x39, auctionSoldBuyNow
0x3a, auctionWon*/auctionLost*) but NO deserializer consumes it - every
candidate function was checked for the value-SKIP/atom-loop signature and none
qualifies. Server-side or telemetry only; it does not carry sold state here.
TASK B IS UNDECIDABLE FROM THE CLIENT, and this is a proof of absence: no 0.95 or
0.05 constant of either width, no tax/fee/net/proceeds caption, and no fee
arithmetic anywhere. The client never computes or displays a net, so no experiment
against our own server can measure the rounding - whatever we credit is what it
displays, and there is no oracle. Only an original EA-era seller-balance capture
could settle it. The rule stays an explicit CHOICE (floor the fee, so
fee + proceeds == gross exactly) and is now pinned at the requested boundaries
100/101/119/120/149/150/151/199/200 plus 15,000 and i64::MAX.
Settlement NOT promoted. No production process, port or database was touched.
Return to Club is durable across a full FUT exit/re-entry — the last claim the
forSale promotion could only make server-side.
Same disposable card as Gap 1 (75 ST, res 212188, wire 100000178, tradeId
1000000178), after the 1h auction expired NATURALLY. No timestamp was mutated in
either gap; a read-only sampler watched the whole hour (55 samples), because
`expires` is derived from created_at + duration and watching is the only honest way
to see expiry:
active expires 3175 -> counting down -> expired expires 0, itemState forSale
(absent) total 0 <- Return to Club
itemState stayed forSale across active -> expired, the one state change Fix B had
never been watched through live.
The host log then shows the coupled transition and TWO session boundaries:
route=move-items wire=100000178 pile=club auction_cancelled=1
route=auth-delete
route=auth ... sid_opened=true (fresh session, x2)
route=hub clubPlayers=1966 auctionCount=0
route=club total=1986 emitted=1966
auction_cancelled=1 is cancel_active_for_core_item firing, so pile membership and
auction lifecycle cannot disagree. Two independent fresh sessions each rebuilt the
state from durable storage and the operator confirmed the card was still in My Club;
one boundary was the requirement.
18/18 server checks pass IDENTICALLY before and after re-entry: /tradePile total 0,
counts all zero, tradeId -> closed with expires 0 (still resolves, correctly
terminal), /club 1966 with itemState free, 0 duplicate ids, market store cancelled
with 0 active and 0 reserved, coins 29,843,976 unchanged throughout.
No code change was required for EITHER gap. Both tests existed to find out whether
the promoted implementation was already correct on paths it had not been exercised
on, and it was.
Also freezes the full CLUB -> list -> active -> expired -> Return to Club -> CLUB
lifecycle as the reference baseline, with the eight invariants it pins, so a future
change that alters any line is a regression until proven otherwise. Explicitly NOT
established: the SOLD path, /tradePile/counts semantics, the AVM1 gate.
Quick-selling a card that was in any squad failed with SQLite 787 FOREIGN KEY
constraint failed, on the live FIFA 17 path (economy_store -> econ.sell_item ->
POST /economy/sell-item). Reproduced, then fixed by evicting the item from every
lineup inside sell_item's existing transaction. routes/cards.rs::delete_owned_card
folded into the same authority, which also gives it the transaction it never had.
Not deployed.
Core gains the generic settlement (gitlink 31ab4a6); the FIFA-specific parts live
here.
FEE (openfut-adapter-fifa17/src/fut/economy_policy.rs), beside pack_price and
match_reward_total because 5% is a game policy constant and Core must stay
game-neutral — Core only validates 0 <= fee <= gross and never computes a rate:
TRANSFER_MARKET_FEE_PERCENT = 5
transfer_market_fee(gross) = floor(gross * 5 / 100), i128 intermediate
seller_proceeds(gross) = gross - fee
Integer only. Floating point is never used for coin settlement: 0.05 is not
representable in binary and a f64 round trip can create or destroy a coin at large
prices. Widening to i128 makes overflow unreachable for any i64 price, so no price
ceiling has to be assumed.
ROUNDING IS A CHOICE AND IT IS NOT CONFIRMED. The fee is floored, so the seller
keeps the fractional coin, chosen because it makes fee + proceeds == gross hold
exactly at every input — the property the accounting invariant rests on. The
discriminating case against flooring the seller's 95% instead is a gross of 150:
this rule pays 143, the alternative 142. Nothing in the corpus or the client binary
settles which the real server did (the client is only ever told the gross; no
tax/netPrice/sellerProceeds wire field exists). Pinned at 0/1/19/20/21/39/40/100/
150/200/1_000/15_000/15_000_000/i64::MAX plus a fee+proceeds==gross sweep.
HOST: CoreEconomy gains settle_sale + EconomySale/EconomySaleReceipt, implemented on
HttpCoreClient as POST /economy/settle-sale. Request field names were checked
against Core's actual SettleSaleRequest/SaleReceipt rather than assumed. Absent club
ids are OMITTED from the body (not null), which is what Core's Outside/active-club
defaults depend on, so a unit test pins that body shape. handle_market_buy is
deliberately untouched: the synthetic buy path has no counterparty, so minting there
is correct.
HARNESS: scripts/settlement-staging.py, stdlib only, drives a REAL Core over real
HTTP on an ephemeral port against a throwaway DB (production 8099/8199/18080 in a
hard deny-list checked in three places), seeds the canonical two-party fixture,
prints BEFORE/PURCHASE/AFTER with PASS-FAIL lines, cleans up in a finally. 31/31
pass. It found the rejection-precedence bug fixed in Core, and that Core's content
preflight aborts startup on an owned card whose CardDefinitionId no pack defines.
Gates: Core 194, adapter 217, host 127, harness 31/31, clippy clean, new code
fmt-clean. Nothing deployed; no production process, port or database was touched.
The one claim the Fix B promotion left open: no active listing existed during
that session, so only the expired path had been exercised.
Closed with a disposable card (75 ST, res 212188, wire id 100000178 — one of
three identical copies, not in the squad) listed through the real FIFA 17 client
at 150/200 for 1h, so expiry arrives naturally. No timestamp touched.
Wire: itemState=forSale, tradeState=active, 12 atoms, prices intact, expires
3562 -> 3556 over a 6s sample (live clock), /trade/status coherent, counts
{count:1, selling:1}, coins unchanged, zero inactive rows (Fix A intact).
The decisive evidence is client-side, not ours: a read-only /proc/<pid>/mem
decode of the live trade-pile auction record returned
itemState=5(forSale)
on the very field that read -1(<unrecognised>) under listFS. Direct A/B on the
only changed field, taken from the client's own memory.
Operator confirmed the row renders under LISTED ITEMS with correct prices, a
counting-down timer, normal art, and correctly non-actionable while active.
No code change required — the promoted implementation was already correct on the
active path. Claim boundary unchanged: this proves the client DECODES the token
and says nothing about the Flash action-gate term.
Operator drove the expired row 1000000155 (res 158023, 93 RW) in FIFA 17 with the
candidate deployed: the row was still actionable, Return to Club was offered, and
it worked — "the card is back in my club".
Server-verified durable afterwards, which is what a fresh session reconstructs:
/tradePile total 0 with zero rows, /tradePile/counts all zero, the card present in
/club (1965 -> 1966 items, itemState "free"), zero duplicate ids, no stale active
listing anywhere in the store (both rows cancelled), and the ended auction
projecting as `closed` (4) on /trade/status. Fix A intact: zero `inactive` rows.
So `listFS` -> `forSale` is protocol-correct AND behaviour-preserving on the path
that matters, and `listFS` is gone from production serialization.
Also worth recording what the result rules out: CARD_OFFERSTATE is NOT a gate term
that requires -1. Every actionable row we had ever seen carried itemState -1, which
looked like a possible client rule; it was a coincidence of our own invalid token.
An expired row decoding CARD_OFFERSTATE = 5 stayed actionable.
Deliberately NOT claimed: anything about the Flash action gate itself. STATE and
the RESERVEDPRICE/MAX_CREDITS pair are untouched and still confounded, so the gate
remains Category C / STRONGLY SUPPORTED / not proven, and AVM1 disassembly of
tradepile.isInActiveAuction is still the separate next investigation.
One gap left open honestly: no ACTIVE seller row existed during the session, so
active-row rendering under `forSale` is unverified. /transfermarket has always
emitted `forSale` on active rows, so it is expected-safe, but it has not been seen.
Single-field protocol-correctness fix, deployed as a candidate for a live A/B.
`itemData.itemState: "listFS"` on the seller's own auction rows is not a FIFA 17
token at all: zero occurrences in `CardsDLL_Win64_retail.dll` (md5
4de3493131d7d2ff7f8b360c5ac9b655), zero in 4.26 GiB of live client memory, and it
decodes to -1 through `FUN_180166660` — so the client was handed an unrecognised
`CARD_OFFERSTATE`. FIFA 17's value for an item offered for sale is `forSale` (5),
from the 12-row table at 0x180229cc0.
Changed only where the invalid token was emitted: `handle_market_query`
(GET …/tradePile) and `handle_market_status` (GET …/trade/status). The market
search path already emitted `forSale` and is untouched — which is also why the
risk here was lower than it looked: the client has been decoding `forSale` on a
live route all along, and only the seller's own pile carried the bad value.
Wire A/B on the same expired row: EXACTLY one field differs. tradeId, tradeState,
expires, startingBid, buyNowPrice, currentBid, bidState, sellerName,
sellerEstablished, watched, coinsProcessed, the twelve-atom count and the whole
itemData card are byte-identical; coins unchanged at 29,843,976; Fix A's zero
`inactive` rows intact.
The differential asserted PARITY on this field and therefore passed while BOTH
sides were wrong — the exact mechanism by which the defect survived every run.
`market query tradePile` is now DIFFERENT-BY-DESIGN, pinning oracle == "listFS"
and rust == "forSale" so the divergence cannot silently close again. Where the
FIFA 17 binary contradicts the Python oracle, the binary wins.
Gates: 126 host tests, 214 adapter tests, fmt clean, clippy clean.
NOT claimed: that this preserves the list -> expire -> Return-to-Club lifecycle.
That needs an operator FIFA 17 session and has NOT been observed yet. Also not
claimed: anything about the Flash action gate — `CARD_OFFERSTATE` is one of three
still-confounded candidates and this change does not test it. Revert is one line
if the live test fails.
RE of the FUT front-end closed the question the Actions-panel investigation left
open, and the answer retracts Q2 rather than completing it.
`tradeState` reaches exactly ONE native branch in CardsDLL — `cmp …,0x4` at
`0x18013e619`, "is it closed?" — and `inactive`(2) and `expired`(3) take the same
edge, producing bit-identical `flagA`/`flagB` (exhaustive 22-site census of
`[reg+0x88]` reads across the PE; confirmed live, both classes read glow=0
inbox=0). The value is then handed to the movie verbatim as the Flash property
`STATE`, and the action gate lives in the APT/ActionScript FUT front-end: the
trade-pile class partitions rows with `getCardsInAuction`/`isInActiveAuction`
(traces `initPile() - IN AUCTION:` / `- NOT IN AUCTION:`) and only auction rows
reach `PreCheckCardOptions` -> `handleTradeCardAction`. A non-auction row renders
and can never be acted on, which is exactly what the operator saw.
So the rows were never usable. "LIVE-CONFIRMED" established that they RENDER,
which is not the same claim, and I treated it as if it were.
The corpus said this before any of it was built —
`plan-2026-08-06-transfer-market.md:731-733`: "`inactive` decodes but no client
path treats it specially; do not emit it." The earlier note explaining that the
warning "was written about the PRESENTATION function" was motivated reasoning.
This also fires the corpus's own pre-registered falsifier E3 (:368-373).
Removed: the `inactive` projection from `GET …/tradePile` and `…/trade/status`,
`UnlistedCandidate`, `resolve_unlisted_pile`, `unlisted_record`,
`Server::resolve_trade_pile`, and the two helpers that existed only to feed them
(`MarketStore::blocking_core_items`, `Fifa17IdentityResolver::wire_for_owned_id`).
Unlisted trade-pile membership is now internal state with no wire expression.
Nothing is stranded: `/club` excludes only items with an ACTIVE listing, so an
unlisted pile member stays visible in the club, which is where the client can act
on it. Verified live after deploy — `/tradePile` total 7 -> 1 with zero `inactive`
rows, `/trade/status` resolving only the real auction, coins unchanged at
29,843,976, and all six former rows present in `/club` (1965 items).
Tests: 126 pass, fmt + clippy clean. Two guards replace the three tests that
pinned the old behaviour: `the_trade_pile_advertises_only_real_auctions` and
`trade_status_answers_only_about_real_auctions`.
NOT fixed here, deliberately: `itemData.itemState: "listFS"` is not a FIFA 17
token (0 occurrences in CardsDLL md5 4de3493131d7d2ff7f8b360c5ac9b655, 0 in
4.26 GiB of process memory, decodes to -1; the real value is `forSale` = 5, and
the Python oracle emits `listFS` too — which is why the differential never caught
it). `CARD_OFFERSTATE` is one of three unresolved action-gate candidates and
every actionable row observed carried -1, so that change ships alone with its own
live A/B.
Normal users press Launch FIFA 17; LSX, autopatch, client preparation and the
pre-launch checks are orchestrated automatically, reusing whatever is already
healthy, and every manual control moves under Advanced / Diagnostics. Service
ownership is tracked so a service the launcher did not start is never killed.
Bumps openfut-launcher past two test-hygiene fixes found by running the suite on
the game machine (.105) instead of only on the server host: the new hook-config
check added a second warning on any box with a hook deployed, and the
shadowed-hostname test was asserting, via backend_reachable's live sockets, that
the local machine has the OpenFUT ports open. Suite now passes on both hosts.
Bumps openfut-launcher to 357501f: Welcome/"Get started" onboarding, Settings as
the single owner of the server address, server-authoritative account claiming,
and `openfut.cfg` reconciled before every launch so a settings change can no
longer leave FIFA pointed at the previous server. Cargo.lock picks up
parking_lot for the launcher (already used by openfut-utas-host and
openfut-identity).
Explains and fixes the Phase C partial failure WITHOUT changing a single wire field.
The operator saw a difference between the one-item probe (Time Remaining "-") and the
generalized rows (Time Remaining "Expired"). Cause: route coverage, not encoding.
/tradePile advertised the unlisted tradeIds while ISVIEWTRADE (GET .../trade/status)
resolved ids from the market store only -- and an unlisted pile member has no listing
row, so the poll returned an empty auctionInfo. Observed live as
`route=market-status requested=1 returned=0` repeating for the row the operator had
selected, while that same id was present in /tradePile. The client polls status for
the row it displays and degrades it when the answer is empty, which is also why no
actions were offered. The probe showed "-" only because the client had not yet polled
that id (logs of the time show only tradeIds=1000000097).
So expires, tradeState, itemData.itemState and pile were all innocent. Nothing was
guessed and no field changed: both routes now share one pile enumeration
(Server::resolve_trade_pile), so an id advertised by /tradePile always resolves on
/trade/status. The corpus predicted exactly this -- tradeId must resolve across
/transfermarket, /tradePile, /watchList AND /trade/status; we had stability but not
coverage. Same defect class as the original empty-trade/status bug.
Status still answers only the ids actually asked about, and a real auction always wins
over an inactive row for the same tradeId. Regression test covers all four cases.
Records the downgraded conclusion: "inactive" is a CONFIRMED section/lifecycle
discriminator; whether the full actionable contract is now complete is the operator's
next test. itemState/pile recovery was queued on the assumption the encoding was
incomplete -- neither was touched, and both remain the next candidates if actions are
still absent.
342 tests pass, 0 failed, clippy clean. Verified live: the six inactive ids went from
returned=0 to returned=6.
Q2 is LIVE-CONFIRMED (operator saw the inactive row under TRANSFER LIST with Start
Price 0 and no Buy Now / Current Bid / timer, active rows still separate under LISTED
ITEMS, and the state survived a full FUT exit/re-entry). Promoting from the bounded
one-item probe to the real behaviour: the env gate is gone and /tradePile now
enumerates the whole trade pile.
Mechanism: read the pile (async), resolve each member to a shaped card (sync, because
the identity/Core resolvers are not `Send`), then build the response (async). The
core->wire lookup is `wire_for_owned_id`, which uses the identity store's
NON-allocating `external_for` -- enumerating a pile is a READ and must never mint a
wire id for an item the client has not seen. Items with no mapping, no Core record or
no resolvable FIFA identity are skipped, never faked.
Includes a bug the DIFFERENTIAL caught and unit tests did not: a pile row OUTLIVES its
auction, so after a sale the seller's `trade` row is stale, and filtering only on
ACTIVE listings re-advertised a SOLD card as an owned unlisted item. Suppression is now
by listing state via `blocking_core_items()` -- active (real auction shown instead),
reserved (sale in flight) and sold (card gone) -- while `cancelled` is deliberately NOT
suppressed, because a cancelled listing means the card came back to the pile. New test
covers all three plus the store-level rule.
counts semantics deliberately unchanged: `count`/`selling` still track auctions only.
341 tests pass, 0 failed, clippy clean. Deployed: the 6 previously stranded pile items
now render, alongside the 1 active listing, with Ronaldo correctly in /club and out of
the pile. Body preserved as phase-c-full-pile-exposed.json.
Operator confirmed in the real client that a tradeState "inactive" row lands under the
right-hand TRANSFER LIST section and renders Start Price 0 with Buy Now, Current Bid
and Time Remaining all absent, while an active row in the same body continued to
render separately under LISTED ITEMS. That is the Q2 representation confirmed live,
with the token itself dumped from the client's own string table rather than guessed.
Records the observed wire->UI contract as a table plus a fixture
(inactive-row-live-confirmed.json), and names the regression tests that pin it,
including the two guards that the row is never emitted for an item outside the trade
pile and never duplicates a real auction.
Also records the expired->Club coupled transition verified server-side: the listing
went to `cancelled`, the pile went to `club`, /tradePile dropped the item, /club
regained it, and clubPlayers went 1964 -> 1965. That is durable store state rather
than a client-local view, so it survives a session boundary by construction; tagged
pending the operator's final exit/re-enter confirmation.
Adds the counts observation table. `count` currently tracks AUCTION entries and not
total Transfer List membership; semantics deliberately left unchanged until the full
state set has been observed.
No behaviour change in this commit.
PHASE A settled the token from the CLIENT ITSELF, so this is not a guessed enum.
vocab_dump.py (new; static, read-only, VA->offset through the real PE section table)
dumps CardsDLL's NULL-terminated {const char*, int} vocabularies. The tradeState
table at 0x180229e40 reads exactly:
'active' = 1 'inactive' = 2 'expired' = 3 'closed' = 4
The sibling tables (type/zone/lev/pos) match the corpus verbatim, which validates the
dumper. So "inactive" is a token the client's own parser decodes.
PHASE B, bounded as instructed. `OPENFUT_FIFA17_UNLISTED_PROBE=<wire id>` exposes
EXACTLY ONE unlisted trade-pile item on /tradePile as a non-active record; unset,
behaviour is byte-identical to before. The other stranded pile items are untouched --
no bulk migration.
Why this shape is forced rather than chosen: the route table has exactly one
trade-pile route, it carries only twelve-atom auction records, `pile` (0x226) has no
deserializer arm so membership comes from the owning list, and of those atoms only
tradeState expresses lifecycle. The row carries tradeState "inactive" with
expires/prices/bid all zero so it cannot render a countdown or a price, and reuses the
item's stable tradeId because the client keys its record store on tradeId and
re-parents itemData -- so listing the item later UPDATES the row instead of leaving a
duplicate ghost.
Both preconditions are re-checked at response time: the item must actually be in the
`trade` pile, and it must not already own a listing. Two tests cover exactly those.
counts semantics deliberately unchanged -- the inactive row is not counted.
340 tests pass, 0 failed, clippy clean. Deployed; the wire now carries all three
lifecycle states at once (expired 1000000097, active 1000000155, inactive 1000000059)
and that body is preserved as a fixture.
Re-entry discriminator came back a CONFIRMED BUG: an unlisted transfer-list item does
not survive a fresh FUT session, so our representation cannot reconstruct trade-pile
membership. Evidence acquisition per instruction, corpus and PE first, no guessing.
Adds route_table_dump.py: static read-only dump of CardsDLL's route table from the
on-disk PE, resolving VA->file offset through the real section table instead of
assuming a single .text mapping. Output preserved as evidence. It settles "is
/tradePile the only relevant route?" -- the table holds 45 routes plus 3 empty admin
slots, and row 30 `ut/%s/tradePile` is the ONLY trade-pile route. There is no
trade-pile items route.
That plus three existing PE facts narrows the representation to exactly one candidate
by ELIMINATION rather than choice: the route carries only twelve-atom auction records;
`pile` (0x226) has no arm in the item deserializer so membership is conferred by the
owning list and cannot be added as a field; of the twelve atoms only tradeState
expresses lifecycle; and tradeState's closed vocabulary (active=1 inactive=2
expired=3 closed=4) has exactly one value not already spoken for.
So an unlisted item can only be an auctionInfo record with tradeState "inactive".
Tagged INFERRED-BY-ELIMINATION, not CONFIRMED: the remaining unknown is whether the
Flash Transfer List RENDERS such a record in the unlisted section. Records the
acceptance test (survive a full FUT reload) and the revised invariant that a
transition is complete only when a fresh session reconstructs the same visible state.
No behaviour change in this commit.
Q2 measured, not guessed. The operator moved a card Club -> Transfer List without
listing it (PUT /item, no POST /auctionhouse) and the state was captured read-only.
Finding: we do not represent the unlisted state on the wire AT ALL. Such an item is
byte-identical to a club item -- itemState `free`, no `pile` field emitted, still
returned in /club and counted in clubPlayers -- while an actively-listed item is
correctly excluded from /club and present in tradePile. Only the host's own pile
store knows the difference. Trade pile held 6 items: 1 listed, 5 unlisted.
The FIFA 17 ENCODING of that state stays UNKNOWN on purpose: returned itemData.pile
is numeric with an unrecovered mapping, and tradeState is a closed table walk where
an unrecognised bidState is silently swallowed as `none`, so a wrong enum produces a
plausible-looking but wrong UI. The corpus warns `inactive` decodes but no client
path treats it specially. One client-only discriminator is recorded instead.
Also models the domain boundary both limbo bugs came from: pile membership and
auction lifecycle are separate facts requiring coordinated transitions. States the
testable invariant -- an item must never be simultaneously excluded from /club and
absent from /tradePile -- with the two ways it was reachable and the commits that
closed each.
CLOSES the active-own-auction Actions-panel investigation. Live client plus the RE
corpus plus historical FUT behaviour all agree: an active auction is COMMITTED until
sale or expiry and is not seller-actionable, while an expired unsold item becomes
actionable (relist / return to club). Every observation fits that lifecycle --
active+frozen expires was non-selectable, expired was selectable and relisted fine,
relisting made it active and non-selectable again, and the client never emits a
cancel. Documented with confidence tags, and the dead ends are named so they are not
retried: MAY_BE_REMOVED is a constant 1, and the eight-flag array is the CLUB-CARD
menu with no auction-cancellation flag in it.
Implements the return-to-club transition that closure exposes. A pile move to `club`
now cancels any ACTIVE listing on that item, because the auction that put the card
in the pile has to end with it. Otherwise the pile reads `club` while the row stays
`active`, so the card is filtered out of /club (exclusion keys on active listings)
AND still rendered in the Transfer List: the move appears to do nothing. This is the
same limbo class as the earlier pile-vs-listing bug, found by reasoning about the
transition rather than by another live failure.
Scoped to `active` only: a `reserved` row is mid-sale and a `sold` row is already
gone, so cancelling either would let one card be both sold and returned. Two tests
cover exactly that boundary.
338 tests pass, 0 failed, clippy clean.
Records the live-client resolution so the market shape is now a fixture rather than
folklore, and so a future agent cannot burn deployments on tradeOwner again.
Confidence notes updated: tradeOwner remains FIFA17-HISTORICAL (it does exist in
the FIFA 17-era API) but "required by FIFA17.exe Transfer List Actions" is now
DISPROVEN for this client path -- it is not among the twelve atoms the client's
auctionInfo deserializer reads, and it was implemented, deployed, observed inert
and removed.
The real blockers are recorded as CONFIRMED live-client findings: trade/status
polling is load-bearing, `expires` must EVOLVE with wall-clock time (a frozen
value is structurally valid and behaviourally broken), and relist must persist
through the PK conflict that FIFA's re-sent ISStart necessarily causes.
Freezes the known-good bodies under docs/evidence/market-lifecycle-2026-08-17/
with a machine-checkable countdown proof (_index.json._countdown_proof records
expires decrementing, frozen:false) rather than asserting the clock in prose.
States the general rule this cost us: a response can pass differential parity and
render perfectly while still being wrong, because FIFA expects an evolving
server-side state machine, not a static object that resembles one.
The client's relist arrives as a fresh ISStart (`POST /auctionhouse`) for an item
that ALREADY has a listing row, so `create_listing` hit a primary-key conflict. The
handler treated `Err(Conflict)` as success: it logged `listed=true`, handed the
client its trade id, and persisted nothing. The stale row kept its old `created_at`,
so the card stayed expired and the relist appeared to do nothing -- observed live,
with the client's price-limits fetch and the ISStart POST both in the log.
The PK conflict IS the relist path. `relist_listing` now resets `created_at` to now
and takes the new prices and duration, so the auction actually returns to the market
with a fresh countdown.
Refuses to revive a `sold` or `reserved` row: re-opening a sold auction would sell
the same card twice. `cancelled` rows ARE relistable (the card is back in the pile).
Missing rows report NotFound rather than silently succeeding. The failure paths still
ack so the screen cannot wedge, but they now say `relisted=false reason=...` in the
log instead of claiming success.
Three store tests: the clock/price reset, the sold+reserved revival guard (plus the
cancelled-is-relistable case), and NotFound.
336 tests pass, 0 failed, clippy clean.
Adds trade_gate_probe.py (read-only: /proc/<pid>/mem O_RDONLY + pread, slide proven
against the on-disk FNV prologue), extending gate_byte_probe.py to vtable slot
+0x270 exactly as the transfer-market analysis asked for.
Measured: IS_TRADING_ENABLED=1 (was 0 in the Python era), TRADE_PILE_SIZE=100
(was 0), watchListSize=50 (was 0), with four controls reading 1. So every
CardsDLL-supplied input that analysis named as a market blocker is now OPEN, which
the Rust host achieves by construction -- it emits userInfo.feature as {} so the
kill switch at 0x180174f19 never arms, and it already sends pileSizeClientData
keys 2 and 4.
This narrows the Actions-panel question to the exe-side UI script term, and rules
out ownership fields, the gate bytes, the cancel route and the state vocabularies
as candidates -- each on measured or PE-derived evidence rather than inference.
Corrects the record against the CLIENT BINARY rather than library hearsay, using
the project's own reverse-engineering record
(fifa17-recon/docs/plan-2026-08-06-transfer-market.md, read out of the on-disk PE).
REVERTED (refuted): `tradeOwner`, `sellerId`, `offers`. FIFA 17's auctionInfo
deserializer (0x18013e410) reads exactly TWELVE atoms -- bidState, buyNowPrice,
currentBid, expires, itemData, sellerEstablished, sellerName, startingBid,
coinsProcessed, tradeId, tradeState, watched -- and value-SKIPs everything else at
0x180135ff0. Those three fields were added last commit on the strength of
contemporaneous FIFA 17 libraries; the PE says the client never reads them, so they
were inert and could not have been the Actions-panel gate. A preservation emulator
must not emit fields the client does not consume. New test pins the exact set.
ADDED: the auction clock. `expires` is SECONDS REMAINING (never an epoch) and the
client renders a LIVE COUNTDOWN it expects to reach 0. We hardcoded 3600, so no
auction ever aged or ran out. Now `duration` is taken from the ISStart body
(additive `duration_secs` column, defaulting to 3600) and `expires` is derived from
created_at + duration - now, clamped at 0. An active listing whose clock has run
out projects as `expired`/`none`/`expires: 0` -- FIFA 17's relistable state, per the
lifecycle table (active=1 inactive=2 expired=3 closed=4; none=0 outbid=1 highest=2
buyNow=3, both closed vocabularies). Pure projection: no row is mutated, so no
sweeper and no race with the economy.
ADDED: `duplicateItemIdList: []` on GetTradePile, which shares one deserializer
(0x18013e7f0) with ISSearch/ISWatchList over four members and we were omitting one.
CONFIRMED by the same source, so kept: `GET ut/{ns}/trade/status?tradeIds=a,b,c` is
real (ISVIEWTRADE) and my handler matches it exactly, including the comma list.
`ISREMOVETRADE` is `DELETE ut/delete/{ns}/trade/{tradeId}` -- our ORIGINAL spelling
was right. The plain-DELETE arm stays because the same source advises dispatching
on path and being method-agnostic (HTTP verbs are not statically recoverable).
Differential returns to strict key-set parity, with a comment recording WHY parity
is not sufficient: a field absent from both sides is invisible to it.
333 tests pass, 0 failed, clippy clean. Verified live: the twelve-atom record, the
four-member envelope, and the listing correctly reading expires=0 / expired after
aging past its hour.
Records the auction-record field set, route spellings, pile encoding and the four
open UNKNOWNs so future agents neither reopen settled questions nor re-guess enum
values. Each claim tagged CONFIRMED / FIFA17-HISTORICAL / INFERRED / UNKNOWN.
Captures the key methodological lesson: oracle parity is necessary but NOT
sufficient for a flow the oracle itself never served -- our auction record matched
the oracle key-for-key while both omitted the FIFA 17 ownership fields.
Three defects behind "selecting my own Transfer List listing opens no dialog".
Pressing the card emits NO HTTP at all, so the gate is a field in what we already
return -- the client decides locally from the auction record.
1. OWNERSHIP FIELDS (FIFA17-HISTORICAL). FIFA 17 auctionInfo carries `tradeOwner`
(bool), `sellerId` and `offers`; we emitted none of them. `tradeOwner` is the
purpose-built "this auction is mine" flag, and without it the Transfer List has
nothing to key owner actions (Remove / Re-list) on. `sellerId` now carries the
configured persona so it agrees with `tradeOwner` and `sellerName` instead of
telling three different stories. Persona is threaded from config, never baked in.
2. `GET …/trade/status` ANSWERED EMPTY (CONFIRMED from our own live logs). The
Transfer List polls this continuously to refresh live auction state. The tail has
no numeric id, so it fell through `t.starts_with("trade")` into the buy/view arm,
where `trade_id_from_path` fails and the reply is `{"auctionInfo": []}`. The
client asked for the state of its own listings and was repeatedly told there was
none. Now a real handler: `tradeIds` filter, or the whole active pile unfiltered;
unknown ids are absent rather than an error, so a poll never fails closed.
3. PLAIN `DELETE …/trade/<id>` WAS A SILENT NO-OP. Contemporaneous FIFA 17 clients
cancel via `DELETE /ut/game/<sku>/trade/<id>`; only the oracle's
`/ut/delete/game/…` spelling mapped to MarketCancel, so the plain form landed in
the buy/view arm and "cancelled" nothing while returning 200. Both spellings now
map to MarketCancel. Kept the oracle spelling: the differential exercises it.
Why the differential missed all of this: our record's key set was IDENTICAL to the
oracle's, so parity was green. The oracle omits the ownership fields too, because
its own remove flow was never driven by a real client either. The differential now
asserts we COVER every oracle key and that our extra keys are EXACTLY
{offers, sellerId, tradeOwner} -- so an unexplained new divergence still fails,
while the deliberate superset is pinned.
Deliberately NOT changed (no evidence): itemState stays "listFS", expires stays
3600 seconds-remaining, bidState stays "none" for active/unbid, counts stays
count=1, and no FIFA 18+ price fields were added.
332 tests pass, 0 failed, clippy clean. Deployed and verified live: tradeOwner=true
sellerId=33068179 sellerName='CAGE' offers=0 on /tradePile AND /trade/status
(filtered and unfiltered).
A card listed on the Transfer Market rendered correctly in the Transfer List but
pressing it opened NO Actions panel, so Remove / Re-list were unreachable. The one
field where our auction record diverged from the oracle was the seller: we stamped
"EASFC" while the oracle stamps the account's persona name. `fut_account.py`
annotates that very property as "Blaze PDTL.DSNM / LSX GetProfileResponse Persona /
UTAS sellerName", so EA's house name on the player's OWN listing is simply wrong,
whether or not it proves to be the gate on the Actions panel.
Introduces `non_economy::PERSONA_DISPLAY_NAME` as the single source of truth and
uses it both for the `account/sync` default (previously a bare "CAGE" literal) and
as the market seller. Every listing in this store is the player's own -- there is no
NPC seller in a single-account emulator -- so the fallback is the player.
Also strengthens the differential: it compared only auctionInfo LENGTH and
tradeState, so it was structurally blind to this. It now compares the record key
set and each shared field against the live Python oracle, asserts the seller is the
persona rather than EA, and asserts itemData is the full card rather than a stub.
That strengthened comparison passes against the real oracle subprocess, which
establishes two things: our record's key set is IDENTICAL to the oracle's (we are
missing no field relative to it), and sellerName was the only divergence.
NOTE the limit of that evidence: the oracle's own Transfer List remove flow has
never been confirmed against a real client either (the only live datapoint is a
counts-tile bug), so parity is necessary but may not be sufficient. If the client
still offers no dialog, the missing field is missing on BOTH sides and must come
from client instrumentation, not from the oracle.
14 targets green, clippy clean. Deployed and verified live: sellerName='CAGE',
listing intact, coins unchanged.
Operator-supplied research document (authored outside this repo) describing the
player-visible FUT hub state machine. Stored verbatim so it cannot drift, with a
provenance header pinning its standing: it is a BEHAVIOUR target, never a protocol
reference. Its own §43 already forbids inventing route/field/sentinel/empty-state
details from it, which matches project policy (guessing wire spellings is the
documented client-freeze class).
Appended a repo-grounded cross-check that tags each relevant claim CONFIRMED /
CONFLICT / GAP / UNVERIFIED, so a future agent cannot mistake the aspirational
parts for observed behaviour. Notably it CONFLICTS with the recovered client
tables twice (Manager League is deliberately excluded from the consumable
overlay; there is no apply-consumable endpoint upstream at all), and it usefully
confirms that "sent to the Transfer List but not currently listed" is a real FUT
state -- which is exactly the limbo f2c4927 worked around.
Keying the club exclusion on the `trade` pile put cards in limbo: the pile can
hold cards with no active listing (a bare "Place on Transfer Market" move, or a
listing later cancelled/sold), and `/tradePile` renders ONLY active listings — so
those cards were invisible in BOTH views. Live prod had 5 trade-pile rows but 1
active listing, so 4 owned cards had no reachable screen (clubPlayers 1966->1961).
Key on the ACTIVE LISTING instead (market store `core_item_id` of `state=active`).
This is self-healing: the moment a listing stops being active the card is back in
the club, with no extra transition to maintain and no need to invent an
"unlisted transfer-list" wire shape (`tradeState` has no verified spelling for
that state, and guessing enum spellings is the documented client-freeze class).
A bare pile move therefore no longer hides a card. That is deliberate: our
`/tradePile` shows only active listings, so hiding on the move alone would
reintroduce the limbo it is meant to prevent.
Verified live: clubPlayers 1961 -> 1965 (exactly the one listed card hidden, the
4 stranded cards recovered); listed wire still absent from /club; counts and
tradePile unchanged. 14 targets green + clippy clean.
Four defects found by driving a real FIFA 17 client. Each was independently
sufficient to break listing, so all four had to go:
1. Every owned card was shaped `untradeable: true` (adapter item.rs), so the
client greyed out "Place/List on Transfer Market" for the whole club. Owned
and pack-pulled cards are TRADEABLE in FIFA 17; the oracle forces this off
for owned copies too (item_def keeps `true`; instances do not).
2. `POST /auctionhouse` required `itemData.resourceId`, which the client's
FutISStart body never sends (the oracle lists by wire id ALONE). Missing it,
the handler fail-closed and returned 200 while persisting NOTHING. It now
resolves server-side: wire id -> Core owned instance -> its card_id (minted on
a synthetic buy) + FIFA resourceId (the auction record). This also enforces
that a listing can only name a card the club actually owns.
3. An auction record's `itemData` was a 4-field STUB, so the Transfer List had a
row the client could not draw -> "1 item listed" but no visible sale. A
listing now persists a full shaped-card SNAPSHOT (new `listings.item_json`,
additive migration) built by the same `shape_item` shaper `/club` and the
squad projection use, so the auction card renders identically to the club
card. The seller's own pile stamps `itemState: listFS`; market search keeps
`forSale` (the oracle distinguishes these).
4. `/tradePile/counts` shared a handler with `/tradePile`. They are DIFFERENT
deserializers: `/counts` is FutGetAuctionCount, five scalar ints
(count/maxAuctionsAllowed/offered/selling/sold) that it reads and skips
everything else. Served the `auctionInfo` body it left every count at 0, so
the Transfer List screen showed no active sale while the hub tile showed one.
New Route::MarketCounts, classified BEFORE the base tradePile matcher (which
also accepts the /counts path).
Also: a listed card no longer appears in the club. `/club` and the hub's
`clubPlayers` now exclude the transfer pile. Pile membership is host-owned state
Core cannot filter on, so when anything is hidden `/club` reuses the existing
local-filter path (the one `rare=SP` already needed) and paginates the
club-visible set -- letting Core paginate would return short pages. With nothing
hidden the fast Core-paginated path is untouched, and only an EXPLICIT non-club
pile hides a card, so no-pile-row items still default to the club.
Fixed 5 pre-existing test fixtures across 4 targets that listed FABRICATED wire
ids -- only "valid" because the old handler skipped the ownership check.
Tests: 14 targets green + clippy clean, incl. new coverage for the 5-int tally
(asserting it must NOT carry auctionInfo), the full-card snapshot + listFS, and
club pile-exclusion with full-width pagination. The differential test against the
live Python oracle passes.
Verified live on prod: listed=true with a 21-field snapshot; counts
{count:1,selling:1,maxAuctionsAllowed:100}; tradePile renders the 94-rated card;
clubPlayers 1966 -> 1961 (exactly the 5 trade-pile items); listed wire absent
from the club page. Operator confirmed the card is visible in the Transfer List.
Two more real client-hit reads move off the Python proxy:
- GET /item/resource, /defid (Route::ItemDefs): build {itemData:[item_def…]}
for every >=3-digit id in the query, replicating the oracle's item_def
(assetId = resourceId & 0xffffff; hardcoded Ronaldo asset 20801 + a generic
"Player" 75 CM placeholder). The client renders the real card from its local
DB, so the placeholder is exact parity.
- GET /marketdata (+ /marketdata/pricelimits) (Route::MarketData): suggested
pricing, constant band 150..15000. /pricelimits returns a BARE ARRAY (one
{defId,minPrice,maxPrice} per queried defId); plain /marketdata returns an
OBJECT {minPrice,maxPrice}. The container type is load-bearing — object-where-
array froze a live client at the listing screen, so the handler picks it from
the path.
Adds extract_long_ints / extract_defid_param query parsers, shape+parser unit
tests (incl. the freeze-critical container-type assertions), and classify-table
coverage. Deployed to prod-host 2026-08-17; verified owner=RUST 200 for all four
(Ronaldo/placeholder resolve, pricelimits=array, marketdata=object).
Docs: PRODUCTION_AUTHORITY_MATRIX + PYTHON_RETIREMENT_PLAN updated. Remaining
Python tail is now only mutation (user/club), no-Core-model (squad/<n>), and
unimplemented modes (draft/leaderboards/sbs).
FUT modes (Seasons/Tournaments/FUT Champions) and the club-identity service are
disabled in this emulator, so these GET reads return {} verbatim from the Python
oracle. Serve them directly from Rust via a new Route::FeatureOffEmpty +
non_economy::feature_off_body() -> {} (byte-identical to the flag-off oracle),
reducing the proxied Python surface.
- The mutating club rename (user/club) stays on Python (needs a Core write).
- Enabling a mode later requires a real Rust handler here, never a Python
fallback (no split authority).
- classify tests: 5 routes owned + user/club/wrong-method lookalikes stay
Passthrough; updated the stale clubUser assertion.
- Deployed to prod-host 2026-08-17; verified owner=RUST 200 {} for all five.
Docs: PRODUCTION_AUTHORITY_MATRIX + PYTHON_RETIREMENT_PLAN updated.
Host/adapter (deployed to prod-host):
- POST /ut/auth (+/ut/delete/auth): Rust mints sid, opens Rust session, adopts
persona from body; POST /openfut/account/sync full Rust envelope.
- GET /userMassInfo: full Rust (was proxy+overlay), shared build_user_mass_info.
- GET/PUT /clientdata/<key>: new clientdata_store.rs (JSON-persisted).
- GET /club/stats/{country,league,team}: context-aware club_stats_body
(nation/league/team buckets).
- GET /store,/match/keepalive,/captcha,/tfa,/livemessage,/activeMessage: StaticAck.
- GET /watchList, /squad/0, /user: Rust handlers.
- host_test.rs updated for the new routing.
Launcher: bump gitlink to c277213 (shareholder-grade redesign + live account panel).
Docs: PRODUCTION_AUTHORITY_MATRIX, PYTHON_RETIREMENT_PLAN, MATCH_LIFECYCLE, and
route-shapes-2026-08-17 reference fixtures for the still-Python tail.
Fix extractor truncation (bound each HTTP message by Content-Length): userMassInfo
(8 KB) and purchasegroup responses are now complete in the committed corpus, not
cut at 4 KB. Document the userMassInfo envelope contract (target shape for a future
full-Rust migration; needs the clubAbbr/established account triad Core lacks).
Rebuilds the primary-capture corpus lost to .gitignore (Known Issues #200): 10
real-client requests + 12 responses captured during the post-P1 staging A/B on a
real FIFA 17 client, sanitised (SID/authCode/deviceId/MAC/tokens redacted; raw pcap
withheld). Documents the account/sync, empty-My-Packs 65534 sentinel, and userMassInfo
contracts, incl. the finding that account/sync is coupled to Python active-profile
selection (so it can't migrate standalone from the userMassInfo hybrid).
Both files documented a wrong Fire2 header layout as authoritative, the reader
trap called out in Known Issues:
- heat2.py's module docstring labelled its >IHHHHB3s header 'VALIDATED'. The
round-trip only validates the payload length + TDF body; decode->encode with the
same mislabelled header trivially reproduces the capture, so it never tested the
[10:16] field boundaries. Marked superseded; cite the proven layout; warn at
build_fire2_frame. Code unchanged (dead tooling).
- fifa-blaze frame.rs: see submodule commit f4f3396.
Bumps fifa-blaze submodule eccd46f -> f4f3396 (FIFA23 stub; not in the prod
container; no prod impact).
A reachability probe (TcpStream::connect then drop; the launcher preflight makes
them) reaches a TLS acceptor as 'unexpected EOF' — byte-identical to the
certificate mismatch that cost three live gates. The redirector classified the
opening before the acceptor to keep a benign probe from forging a TLS fault, but
the roster host (the second FIFA-facing TLS host) did not, so the documented
hazard 'remains in any other TLS host that has not adopted it' was live there.
Lift the pure policy (PeerOpening + classify_opening) plus a peer_opening(&TcpStream)
peek helper into the shared openfut-tls crate (game-independent; +unit tests).
The redirector now re-exports them (public API + its probe_classification test
unchanged; behaviour identical). The roster host adopts them: a ProbeCount, a
probes() handle, and a pre-acceptor peek that logs PROBE and returns instead of
failing the handshake. New roster probe_classification integration test (3 cases:
bare probe classified, real client after a probe still served 200, speaks-then-
fails still reported as a fault). Full workspace tests green; clippy -D clean.
openfut-hook (Windows version.dll injected into the FIFA client) declared a
[profile.release] with panic=abort/strip/opt-level=s that Cargo silently ignored
because it was a non-root workspace member (per-package `panic` overrides are
forbidden). Move it out of `members` into `exclude`; the submodule now carries a
matching empty [workspace] table so it builds as its own root. Fixes cross-FFI
panic-unwind UB in the injected DLL, shrinks it 1200126 -> 861696 B, and lands the
artifact in openfut-hook/target/ (matching launcher config.rs hook_dll_path).
Bumps openfut-launcher submodule ca7ce26 -> 0d3f33c.
Global MY-CLUB stat set now Rust (staging-verified RUST route=club-stats
owned=1982 players=1962); only club/stats/{country,league,team} nation-bucket
context sub-screens remain proxied (low value, inert atoms).
Migrate the MY CLUB stat set from the Python proxy to a Rust handler computing
Core-accurate counts: player tiers + rare from the collection, staff/consumable
families from catalog kind+subtype, per-nation buckets via the reverse entity
resolver. Faithful port of fut_club_stats.py (VOCAB + global_counts +
context_rows). Unlike the oracle (stale profile + synthetic consumable shelf),
this reflects the real imported content (incl. the content-gap consumables/staff).
Fail-closed 503 on Core error. club/stats/country|league|team sub-screens remain
Python (documented). Adds adapter club_stats module (5 tests), host handler +
classify arm + resolver subtype_of/rareflag_of, ownership + integration tests;
reachability tool splits club/stats global(migrated) vs context(residual).
Migrate GET /hub from the Python proxy to a Rust handler deriving counts from
authoritative state: clubPlayers = owned PLAYER cards in Core (consumables/staff
excluded via catalog kind; may be lower than Python's profile count by the
deferred Legend instances = DIFFERENT-BY-DESIGN), auction/tradePile counts from
the durable market store via the async bridge. Fail-closed 503 on Core error;
market read failure degrades cosmetic counts to 0. Adds classify arm, handler,
ownership + integration tests; reachability tool marks hub migrated.
Migrate GET club/stats/staff from the Python proxy to a Rust static handler.
The production oracle returns {} for the staff-bonus stat set (deliberately
empty); the Rust host now owns it (owner=RUST route=club-stats-staff). Updates
classify(), the pre-existing near-miss test (now club/stats/year), the ownership
matrix test, and the no-fallback integration test. club/stats/{year,consumables}
remain Python (aggregation) pending the Core-derived club-stats migration.
Through real handle_with_ip dispatch against live Core:
- pure_economy_routes gains the retail v2 Store family (transaction/0,
purchasegroup, purchased GET/POST) so NEVER-BOTH + no-fallback assert them
Rust-owned (Python proxy count 0) and fail-closed 503 on dead Core.
- Part-7 repro: PUT /ut/v2/game/fifa17/store/transaction/0 returns a Rust
createPackResponse + debits Core (NOT the Python TRANSACTIONCANCEL no-op).
- retail_v2_store_flow_matches_v1_through_dispatch: v1 and v2 BUY of the same
pack debit + mint identically; v2 purchasegroup + reveal Rust-owned.
Live staging (S2) showed the retail FIFA17 client issues the Store family
under /ut/v2/game/<sku>/... (PUT /ut/v2/game/fifa17/store/transaction/0),
which escaped Rust economy authority to Python. Fix classify_economy:
- ut_tail() normalizes both /ut/game/<sku>/ and /ut/v2/game/<sku>/ to the
same tail (generic sku, never hard-coded fifa17); delete family likewise
accepts /ut/v2/delete/game/.
- StoreBuy matches store/transaction and store/transaction/<digits> via a
bounded is_store_transaction_tail (never store/transactions, ...foo, or
.../<id>/extra), mirroring the Python bare /store/transaction regex.
Adds table-driven ut_tail + is_store_transaction_tail + classify_economy v2
unit tests (lib 74).
barrier_never_both_no_fallback_and_stale_reader drives the REAL post-barrier
handle_with_ip against a live in-process Core + a mock Python upstream that
counts every request and answers with a coins=111 marker:
- STALE READER: credits + userMassInfo show the Core balance, never 111
(userMassInfo proxies the Python envelope but the Rust economy overlay wins).
- NEVER BOTH (Core up): every pure economy route returns a Rust body (no
__python__ marker) and the Python proxy call-count stays 0.
- NO FALLBACK: a server pointed at a dead Core port (built without probing Core:
empty catalog + empty pool) fails closed (credits/match -> 503) and STILL never
proxies to Python (call-count unchanged).
Also parametrizes build_econ_server's pass URL so the mock upstream can be
injected. host 71 lib + 4 integration + differential + concurrency + failure +
24 host_test all green; clippy -D warnings + fmt clean.
The economy authority barrier. `handle_with_ip` now dispatches every
economy-touching route to Rust/Core via `try_handle_economy` BEFORE consulting
`classify()`, so a migrated route can never also reach the Python passthrough
(NEVER BOTH). With economy services wired (production `from_config`) an economy
route ALWAYS returns Some — fail-closed 503 on any Core error — so there is no
Python economy fallback. `userMassInfo` stays a hybrid by design: Python
supplies the non-economy envelope; Rust overlays BOTH the squad and the economy
fields (coins + unopened-pack count from Core), so no stale Python economy value
is visible.
Routing only — no handler/test changes buried here. Routes now Rust-owned:
/user/credits, /store/purchasegroup, /store/transaction, POST+GET /purchased,
PUT /item, item DELETE forms, /match (ut/delete), /auctionhouse, /tradePile,
/trade, ut/delete trade; plus the userMassInfo economy overlay.
openfut-import-fifa17/tests/durable_import.rs drives the REAL import pipeline
(analyze -> Report dry-run; emit_content; plan_apply -> GenericImportRequest +
deterministic identity mappings + watermark; the staging/preflight/seed/
post-validate gates over a real JsonIdentityStore; openfut_core::services::
import::apply_profile_import in one Core SQLite tx) against a disposable
temp-file Core DB, from a small sanitized in-test fixture (750000 coins, 3
resolvable base players, unopenedPackIds [70,70,101], one squad).
Five ordered steps on one durable target, all green:
A dry-run: report exposes persona/coins/inventory/unopened/fingerprint; ZERO
DB mutation (all Core tables COUNT=0, identity store empty).
B apply: coins=750000 exact, owned=3, packs=3 (opened=0), squad_players=2,
deterministic owned ids, import_fingerprint recorded, identities reverse-
resolve both ways, watermark=100000600.
C restart: close+reopen the SAME sqlite file -> identical state.
D re-apply same source -> AlreadyImported (fingerprint), no doubling.
E conflict (coins 750000->750001 flips the fingerprint for the same game)
-> apply fails closed ('different source'); DB unchanged.
Fingerprint = FNV-1a-64 hex of the source snapshot, carried into
ProfileImportRequest.source_fingerprint = Core profiles.import_fingerprint, the
per-game rerun-identity key. dev-deps added to openfut-import-fifa17
(openfut-core path, tokio, sqlx). No production/live data.
economy_differential.rs boots the REAL Python oracle (fifa17-recon/tools/
utas_server.py) as an isolated subprocess (env FUT_PROFILE/FUT_ACCOUNT_PATH/
FUT_PORT into a temp dir + loopback port; no production container/port/save;
killed on Drop) AND the real Rust stack (seeded in-process Core + a real Server
with EconomyServices), seeds a semantically-aligned fixture on both, and drives
16 ops through the REAL surfaces (oracle over HTTP; Rust via
Server::try_handle_economy on the off-runtime thread).
Result: 15 PARITY, 1 DIFFERENT-BY-DESIGN.
- PARITY: credits, userMassInfo economy, purchasegroup (pack70/sentinel/clean-v1
incl. the real SessionStore capability handshake), Store BUY, POST /purchased
open, GET /purchased reveal (VERIFIED: durable single-profile purchased pile
on BOTH — the hypothesised per-SID cache does NOT exist, so PARITY not
DIFFERENT-BY-DESIGN), quick-sell (both forms), move, match WIN (+400 byte
shape), market list/query/cancel.
- DIFFERENT-BY-DESIGN: market second-buy. First buy debits buyNowPrice + closes
on both. Rust's MarketStore is a crash-consistent single-debit ledger (second
buy of a sold listing = no-op, pinned by assertion); the oracle's buyable
market is a stateless PACK_POOL sample that re-debits on repeat. Compat impact
NONE (buy-now is one-shot); Rust is a strict correctness improvement.
No Python source changes; no classifier changes. Deterministic (3 runs).
Two real host-dispatch test files (no fakes) driving Server::try_handle_economy
against a live in-process Core over the real blocking client + durable
MarketStore/PileStore + JsonIdentityStore, each racer its own OS thread
(off-runtime pattern).
economy_concurrency.rs — 8 races x 50 iterations:
A two BUYs (coins for one) -> exactly one 200 + one 461, final 0, one debit.
B duplicate owned-pack open -> one redemption, +11 once, entitlement once.
C duplicate quick-sell -> one sell + one credit + one removal.
D two market buyers -> one win, one debit, one mint, sold once.
E reward+BUY -> no lost update (Core relative UPDATE under BEGIN IMMEDIATE).
F move+quick-sell / G list+quick-sell -> one coherent transition.
H 1000 concurrent mints -> unique + reversible wire ids, monotonic watermark.
economy_failure.rs — 10 fault-injection sub-cases, all fail-closed:
BUY/open-redeem/generator/pile/identity, quick-sell, move, market
reserve/purchase/complete. CRITICAL complete-sale-after-commit = SAFE: the
listing is left `reserved` (not active), so the active->reserved reserve CAS
can never win again -> not buyable, exactly one debit + one mint. No E3.
Fault injection uses test-file CoreEconomy/ExternalIdentityStore doubles plus a
NARROW, inert-by-default `StoreFault` seam in market_store.rs + pile_store.rs
(the concrete stores have no trait boundary; 3 `tripped()` checks + a field,
zero behaviour unless a test arms it). `parking_lot` promoted to a normal dep
(the seam's Mutex is used at lib scope). Classifier/ROUTE_AUTHORITY/Python
untouched. host lib 71/71; both new tests pass.
Wire the PRODUCTION constructor so the economy authority is not test-only.
Server::from_config now builds one process-lifetime AsyncBridge, opens the
durable MarketStore + PileStore (paths from config), shares one HttpCoreClient
as both CoreAccess and CoreEconomy, builds the content pool from Core, and
attaches EconomyServices via with_economy. Stores/bridge are host-lifetime, never
per request.
- config.rs: required OPENFUT_MARKET_DB / OPENFUT_PILE_DB (durable file paths;
must survive host restart — no temp defaults).
- Fail-closed startup: a bridge/store that cannot initialize returns Err from
from_config (host refuses to start) — NEVER a silent omission or a Python
economy fallback.
Test: from_config_constructs_and_serves_economy — builds the Server via the REAL
from_config (disposable config: temp market/pile/identity + a catalog file
derived from seeded content + the real tables dir) against a live Core, drives
credits / purchasegroup / Store BUY / market list-query-buy through it, then
rebuilds from the SAME config after a Core restart and asserts the balance
persisted. host 71 lib + 3 integration + 24 host_test green; clippy/fmt clean.
Closes the reveal contract gap: POST /purchased opens a pack and returns
metadata; the client then polls GET /purchased for the opened items. Store BUY
returns items inline, but owned reward-pack (e.g. pack 70) opens had no reveal
read path, so a real FIFA session would show nothing after opening.
Faithful to the Python oracle (fut_store.last_pack / purchased pile): the reveal
is the set of owned items currently in the FIFA "purchased" pile — durable,
idempotent on repeat GET, cleared per-item when a card is moved to the club, and
appended-to by each open. Not a replay cache; presentation state derived from
the durable pile store + Core inventory.
- pile_store.rs: `list_by_pile(pile) -> Vec<core_item_id>` (reveal membership).
- economy_store.rs: `PurchasedPileSink` trait + optional `StoreDeps.purchased`;
handle_store_buy/handle_pack_open record each minted item into the "purchased"
pile. `shape_purchased_reveal` (pure): filter Core inventory to the purchased
pile, shape with the SAME `shape_club_response` /club uses. Grants nothing,
consumes no entitlement, allocates no id, moves no coins.
- lib.rs: EconomyRoute::PackReveal + classify_economy (GET purchased);
BridgedPurchasedSink (records via the runtime bridge from the sync dispatch
thread); dispatch reads the pile async + Core inventory sync + pure-shapes.
Scoping: single fifa17 profile/club (like the Python oracle), so all sessions
share one purchased pile — DIFFERENT-BY-DESIGN vs a per-SID cache, matching the
oracle's single-profile model.
Tests: pile_store::list_by_pile_filters_and_reflects_moves; and the dispatch E2E
now opens pack 70 (entitlement seeded via the Core economy API) and asserts GET
/purchased reveals the opened items and is idempotent on repeat. host 71 lib +
2 integration + 24 host_test green; clippy -D warnings + fmt clean.
Closes the market correctness gap: handle_market_list recorded listing.card_id
from the raw FIFA wire resourceId, so a synthetic buy minted a card_id Core
could not resolve — it survived the immediate response but Core's content
preflight rejected it on reboot.
- catalog.rs: keep the by_resource reverse index (was built then discarded) and
expose `card_id_for_resource(resource_id) -> Option<&str>` — exact reverse of
the card_id->asset catalog, no heuristics, unknown => None.
- lib.rs: `impl MarketCardResolver for Fifa17IdentityResolver` delegates to the
same catalog /club shaping uses; Core never sees a FIFA resource id.
- market_store.rs: listings now carry BOTH `card_id` (authoritative Core content,
what a buy MINTS) and `wire_resource_id` (the FIFA wire id, echoed in the
auction record). New column; create_listing takes both; row/Listing updated.
- market.rs: `MarketCardResolver` trait; handle_market_list resolves resourceId
-> Core card_id and fails closed (persists nothing) on an unmappable resource;
auction_record emits `resourceId` from wire_resource_id. Dispatch passes the
resolver.
Tests: list_unknown_resource_fails_closed_no_listing (B),
list_persists_core_card_and_wire_resource_across_reopen (C), catalog reverse
lookup; and the dispatch E2E now RESTORES the full Core+store restart
(economy_full_sequence_through_dispatch_and_restart) — the synthetic buy mints a
real reverse-mapped card_id, so Core's content preflight passes on reboot (A+D).
market 23 lib + catalog 15 + 2 integration green; clippy -D warnings + fmt clean.
Records the AsyncBridge + classify_economy + try_handle_economy dispatch
(unrouted) and the economy_full_sequence_through_dispatch E2E, the
reqwest-blocking-in-async fix (off_runtime), and narrows Remaining to: Python
differential, host concurrency matrix, failure injection, importer, then the
from_config attachment + classifier barrier + reachability proofs. Flags the
two market-handler gaps (resourceId->card_id mapping; GET /purchased reveal
cache).
Bridges the synchronous thread-per-connection host to the async
transfer-market/pile handlers WITHOUT flipping the classifier. classify()
is untouched; production still proxies every economy route to Python. The
new dispatch is exercised only by the integration harness via
Server::try_handle_economy — handler wiring, not authority cutover.
async_bridge.rs: AsyncBridge owns ONE process-lifetime multi-threaded Tokio
runtime, shared by every connection via Arc. block_on() runs a future from
the sync dispatch thread; if invoked from within an ambient runtime it
offloads onto its own runtime + a std channel instead of panicking
("cannot start a runtime from within a runtime"). 4 unit tests incl. the
nested-runtime-safety case and concurrent multi-thread drivers.
lib.rs: EconomyRoute + classify_economy (mirrors the Python route table:
credits, purchasegroup, store/transaction, purchased, item DELETE/PUT,
ut/delete match/item/trade, auctionhouse/transfermarket, tradePile, trade).
EconomyServices (Core econ transport + durable MarketStore/PileStore + the
bridge + the pack-content pool), attached via Server::with_economy (kept out
of `new`/`from_config` so existing tests build a DB-less Server; production
from_config attachment is the barrier step). Server::try_handle_economy
dispatches: sync handlers (credits/purchasegroup/store-buy/pack-open/
quick-sell/match) inline; async handlers (market list/query/buy/cancel,
move) on the bridge via owned `async move` blocks. build_content_pool
derives the resolvable FIFA∩Core candidate pool from Core content.
market.rs: FIX the load-bearing hazard the FakeEconomy tests missed — the
async market handlers call the BLOCKING reqwest Core client, which panics
(reqwest::blocking::wait::enter) when run while a Tokio runtime is entered.
off_runtime() hops each Core call to a fresh OS thread with no runtime
entered, so blocking is legal. handle_move_items resolver gains `+ Sync`
(future must be Send for the bridge).
tests/economy_integration.rs: economy_full_sequence_through_dispatch drives
the WHOLE cluster through the REAL Server dispatch + bridge against a live
in-process Core (seeded with fifa17 dev content: 100k coins + owned cards),
on a plain OS thread (direct bridge path), over the real blocking
HttpCoreClient — no fakes: Store BUY (pool draw + shape + debit 400 + mint 5),
credits, quick-sell (reverse-resolve + credit), match WIN (+400), market
list->query->buy->query(sold)->second-buy-fails(no double debit), cancel
(cancelled not buyable), move-items, then reopen the durable market/pile
stores from disk (sold + pile persist). Deterministic. start_core_seeded
loads fifa17 dev content so /collection renders real definitions.
Tests: host 68 lib (+4 bridge) + 2 economy_integration + 24 host_test, all
green; adapter unchanged-green; clippy -D warnings + fmt clean.
Update cutover progress: Store BUY/pack-open/quick-sell, market
list/query/cancel/buy + move-items, durable listing/pile stores, and the
pack-content generator are implemented + tested (4d2b8b9) but NOT routed.
Remaining before the single classifier barrier: sync<->async Server wiring +
classify() routes, host<->Core writer E2E, Python differential, host
concurrency matrix, importer restart/idempotency, then the barrier + no-fallback
proofs.
The fresh-DB write-lock race is fixed (core fbb54ea: BEGIN IMMEDIATE writes),
so the E2E+restart harness now uses max_connections=5; 10/10 deterministic.
Complete the transport contract: purchase_items (atomic debit + mint N) on the
CoreEconomy trait + HttpCoreClient (POST /economy/purchase-items) + FakeEconomy.
This is the open-on-buy primitive the Store BUY handler will use (createPackResponse
returns the minted itemList). Fail-closed like the rest of the client.
WAL-establish-once + busy_timeout (core 75b1830) reduced but did not eliminate a
brand-new-DB multi-connection warm-up 'database error'; the E2E+restart harness
stays on a single serialized connection for determinism, and the residual
multi-connection concurrency issue is documented as the remaining Part-T blocker.
Serialize Core access with a single pooled connection (the harness drives Core
sequentially via the blocking client) to avoid a WAL-mode-establishment race
across connections warming up on a brand-new DB file, and raise the readiness
ceiling for heavy parallel test-binary load. 10/10 deterministic. (Fixed
alongside a real Core robustness fix: per-connection pragmas + busy_timeout,
core 0360135.)
Spawn Core (axum) on an ephemeral loopback port backed by a disposable temp-file
SQLite, seed a fifa17 profile via the real Core HTTP API, then drive the HOST's
REAL transport (HttpCoreClient: CoreEconomy) + handlers against it — no fakes:
credits reads Core balance; match-reward writer credits via Core grant_reward;
purchasegroup full-gen renders the owned pack from a Core entitlement (no
sentinel); userMassInfo overlay derives coins from the same Core state
(credits==massinfo==Core invariant). Restart phase reboots Core from the same
on-disk DB and proves coins + entitlements persist. Temp dir + 127.0.0.1:0 only;
no prod DB/ports/containers/.105. dev-deps: openfut-core, tokio, axum.
Carry unopenedPackIds through the importer: Profile model -> Report ->
ApplyPlan. GenericImportRequest now emits entitlements[] (one definition_id
per unopened pack instance, order+duplicates preserved), which Core's import
seeds as unconsumed packs rows in the same transaction. Closes the economy
import gap so a migrated profile's unopened packs become Core entitlements
(feeding purchasegroup/credits/userMassInfo). Idempotency unchanged (import
fingerprint). +1 test; 26 pass, clippy -D warnings clean.
All Core-backed, fail-closed (503, never Python), NOT yet classifier-routed
(coherent barrier pending full cluster + Core seed):
- handle_purchasegroup: full Rust body from Core entitlements + StoreMode via
the oracle-fixture-tested build_purchasegroup (no Python body dependency).
- overlay_massinfo_economy: set userInfo.currencies coins + unopenedPacks
recoveredPacks from Core, preserving all other fields.
- handle_match_end + build_match_reward_body: derive outcome from endReason,
credit via Core grant_reward, oracle-shaped destroy_match_body.
Invariant test: credits == userMassInfo == purchasegroup all read one Core state.
7 new host tests (+ FakeEconomy write methods honor fail flag).
Add CoreEconomy transport (trait + HttpCoreClient impl over Core /economy/*):
balance, entitlements, purchase_entitlement, redeem_entitlement, sell_item,
grant_reward, purchase_item. Fail-closed by contract: any transport/status/parse
error surfaces a controlled error and NEVER falls back to Python (a fallback
would be a second writer).
Add the credits reader vertical: build_credits_body (byte-shape-identical to the
Python oracle: credits + currencies[].funds/finalFunds + optional
unopenedPacks.recoveredPacks) and handle_credits (coins = Core balance,
recoveredPacks = Core entitlement count; 503 fail-closed on Core error). Not yet
classifier-routed: the coins cluster flips as one coherent barrier once every
writer+reader moves together and Core is seeded. FakeEconomy double + 3 tests
(oracle shape, Core-backed read, fail-closed).
Advance openfut-core gitlink to d32dc6e: generic /economy/* HTTP routes over
services::economy, server-side club resolution (game-scoped active profile, no
client-supplied club id), + list_unopened_entitlements reader. This is the
transport the FIFA17 economy cutover binds to. Core matrix 43 lib + 115
integration green; clippy clean; boundary audit clean.
Machine-auditable ownership table for every FIFA17 UTAS route touching the
economy cluster (coins/inventory/entitlements), plus the writer->Core-primitive
map and the single-writer rule. Grounded in the Python economy writer audit:
4 coin-mutation routes (match reward, pack BUY, quick-sell, market buy-now),
synthetic-seller market (no sale-credit/expiry/fee), dead grant_coins/
grant_unopened_pack, no points writer. This is the deployment gate: no proxied
Python route may touch Core-owned state before R1.
Advance openfut-core gitlink to c8269d0, which adds services::economy::purchase_item
(atomic debit + mint) — the generic Core primitive the FIFA17 synthetic-seller
transfer market needs. Evidence: the Python economy audit proved market buy-now
mints a new item with no real counterparty, so debit+mint (not two-party transfer)
is the correct generic model. Core matrix + clippy green.
Advance the openfut-core gitlink 3084a46 -> ee2caa0. This does two things:
1. Reconciliation: moves the canonical Core lineage onto the committed,
validated migration trunk (66c88fb: game-scoped opaque extension,
inventory service, squad-ext routes, content-pack loader, generic
transactional profile-import service). The divergent local Core refactor
that was dirtying the eab522a checkout is preserved verbatim on branch
wip/core-local-development (f70cf44) for separate reconciliation; nothing
is lost.
2. Economy foundation: ee2caa0 adds services::economy, a generic atomic
profile-economy authority (currency/inventory/entitlements over the
existing durable tables, single-transaction compound ops, fail-closed).
The FIFA17 adapter economy engine sits on top of these primitives.
Core builds green; full matrix 41 lib + 111 integration + 16 = all pass;
clippy clean.
Adds openfut-adapter-fifa17 fut::economy — the single-writer FIFA 17 economy engine
the eventual cluster cutover needs: coins + unopened-pack entitlements + owned
inventory + stable item ids, with all-or-nothing transactional mutations faithfully
ported from the Python oracle's fut_store.Store primitives.
Atomic ops: debit (fail-closed), credit, grant_pack/consume_pack (consume-once),
allocate_item_id (unique/monotonic), add_item, and composed transactions buy_pack,
open_pack, quick_sell, market_buy_now, grant_reward. Fail-closed everywhere; the
65534 sentinel can never be bought/granted/opened (defense at the grant primitive,
mirroring grant_unopened_pack rejecting non-catalogue ids). from_fut_profile importer
round-trips coins/unopenedPackIds/items/nextItemId and floors nextItemId past the
highest existing id so re-import cannot mint a duplicate. 10 unit tests (atomicity,
sentinel safety, consume-once, quick-sell, item-id uniqueness, import round-trip).
NOT wired (R3): the live coin balance is one indivisible writer set spanning Store
BUY, pack-open, quick-sell, match rewards AND the transfer market, all in
fut_profile.json; and the generic home (OpenFUT Core) is a preserved-dirty/frozen
submodule. So a safe single-writer cutover cannot be wired yet — this engine +
importer is the coherent prerequisite. No dual-write introduced. No deployment.
Adds openfut-adapter-fifa17 fut::store_catalog — a pure, faithful Rust port of the
Python oracle's PACK_CATALOG + _pack_body + store_catalog assembly at production
flag defaults (FUT_STORE_DISPLAYGROUP=1, GROUPID=0, PRICE_PROBE=0):
- PackDef + PACK_CATALOG (ids 1/5/6/7/70; economy numbers are OpenFUT PLACEHOLDER,
wire shape is oracle-verified; 65534 deliberately absent).
- pack_body() (_pack_body port), sentinel_body() (id-65534 compatibility shim),
build_purchasegroup(unopened_ids, StoreMode) mirroring store_catalog(3627).
- Differential parity: fixtures generated from the Python oracle
(tests/fixtures/purchasegroup_{zero_sentinel,zero_clean,pack70}.json); Rust output
matches semantically (6 tests). Adapter 140 tests, host 24, fmt/clippy clean,
Python A-R oracle green.
PURE wire shaping — NOT wired into the live host. Serving purchasegroup from Rust
requires an authoritative Rust owner of unopenedPackIds, which is blocked on the
economy-authority prerequisite (R3): coins are one shared balance written by many
Python-oracle routes (BUY spend, quick-sell credit, SBC/match/objective rewards)
persisted to fut_profile.json, so no single coin-touching route can move without a
whole-cluster migration. No dual-write introduced; no production deployment.
openfut-utas-host now classifies and owns three routes, wiring the landed
adapter store_session state machine while keeping the Store economy Python's:
- POST /ut/auth: proxy to Python (which mints X-UT-SID, adopts persona, refreshes
save), OBSERVE the returned sid, and open a Rust session bound to peer IP +
configured persona. Account/economy authority stays Python.
- POST /openfut/fifa17/capability: Rust-owned, no proxy — validate + register into
SessionStore (bound/pending/ignored-late); fail-closed 400 on unsupported.
- GET .../store/purchasegroup: proxy to Python for the authoritative economy body,
then overlay ONLY the empty-My-Packs topology from the frozen session mode —
strip the 65534 sentinel for a verified clean-v1 SID, keep it otherwise. Rust
never writes economy state.
Session state (Arc<Mutex<SessionStore>> + monotonic clock) lives on Server; new()
and from_config() initialise it (signatures unchanged). handle() gains a peer-IP
variant (handle_with_ip) threaded from handle_conn. Strict never-both routing is
preserved. Pure helpers (observe_sid, parse_capability_request, overlay_empty_mypacks)
+ classifier are unit-tested; adapter+host tests + Python A-R oracle all pass.
STOP-GATE: Store BUY / coins / unopenedPackIds NOT migrated — Rust has no
authoritative FIFA17 economy-mutation path (Python fut_profile.json is the source;
Core's economy is separate/unwired), so moving BUY would split store authority.
That cluster migration is the remaining R2 gap. No production deployment.
Ports the novel per-session empty-My-Packs capability negotiation — proven live
on staging and currently Python-only (fifa17-recon/tools/utas_server.py) — into
the production Rust FIFA17 adapter as a pure, dependency-free state machine
(openfut-adapter-fifa17 fut::store_session).
It owns: per-login X-UT-SID session table, single-use (ip,persona) launcher
capability hand-off (pending), capability binding (bound/pending/ignored-late),
the once-per-session clean-v1 vs sentinel freeze, TTL reaping, and fail-closed
rules (unknown/expired/ambiguous/late/cross-session/sid-ip-mismatch -> sentinel).
The clock and SID entropy are injected so it is fully unit-testable.
The full Python capability-negotiation matrix A-R is ported as Rust unit tests
(21 pass). Python remains the behavioural oracle. The delicate _pack_body UTAS
wire shaping, /ut/auth persona-adoption, /store/purchasegroup catalogue assembly
and the store BUY path are deliberately NOT ported here (documented gap); wiring
the three routes into openfut-utas-host without splitting store authority is the
remaining bounded slice toward full Rust authority. No production deployment.
Point the launcher gitlink at the reconciled merge ca7ce26
(integration/fifa17-launcher-capability-sbc), which retains BOTH launcher lineages:
- 13339c1 FIFA 17 verified patched-client capability reporting
- 958ff245 openfut-hook SBC request tracing / RE instrumentation
The previously-uncommitted openfut-hook WIP that blocked this move is preserved on the
submodule branch wip/openfut-hook-local (commit 4e44a37) + /tmp/openfut-hook-wip-preserved.patch
(sha256 8e65de2c…) + the untracked server.rs copy. Gitlink-only change; no other superproject
dirt staged. Backend/production unchanged.
Record the overnight launcher-lineage reconciliation (merge ca7ce26 retaining both
feat/launcher-arming 13339c1 and feat/sbc-hook-tracing 958ff24; only src/process.rs
conflict, resolved keep-deleted), the deferred superproject gitlink bump (blocked by
uncommitted openfut-hook WIP overlapping the merged hook content), the validated
deployment-candidate commit tuple + local build artifacts, and the controlled A/B/C
deployment sequence. Production stays P2 active-sentinel until the A/B passes.
Case R makes the F3 session-topology invariant explicit alongside the A-Q matrix:
for a single X-UT-SID the frozen empty-My-Packs mode never flips in either
direction (Sentinel stays Sentinel even if a capability later appears; Clean stays
Clean even if the capability is wiped), while a fresh SID from the same IP decides
independently. Complements F/G/K.
Record the per-IP -> per-session correction: why source-IP-only was unsafe (two
FIFA processes share an IP), the authoritative per-login X-UT-SID key with IP and
persona as auxiliary, the Capability/StoreMode state machine, the single-use
short-TTL launcher->session pending hand-off, activity-based session cleanup, and
the documented fail-closed residual for genuinely simultaneous same-(ip,persona)
logins. Design history is retained; the per-IP prototype is marked superseded.
Harden the empty-My-Packs capability binding so a verified FIFA process can never
enable clean/no-sentinel Store topology for another unverified process that merely
shares its source IP. The prototype keyed the decision by source IP alone; two FIFA
processes (concurrent, or a relaunch) share an IP, so an unpatched process could
inherit a patched one's clean-v1 mode and crash. Source IP is now auxiliary only.
- Authoritative key = the per-login UTAS session id (X-UT-SID). /ut/auth now mints
a fresh unique SID per login (was a shared constant) and opens a session record
keyed by that SID; the client echoes it on every later call incl.
/store/purchasegroup (live-confirmed). The legacy constant is still accepted by
the retired security-question gate only, never to grant clean-v1.
- Session state: _FIFA17_SESSIONS[sid] = {ip, persona, resolver, mode, created,
last_seen}. Store mode freezes at the first /store/purchasegroup of the session
and is immutable thereafter. Fail-closed: unknown SID, or a SID presented from a
different source IP than it was opened on, resolves to the sentinel.
- Launcher capability (out-of-band; cannot know the SID) is matched by (ip, persona)
as a SINGLE-USE, short-TTL pending, bound to exactly one session at whichever comes
first: its login (pending predates auth), the registration (session already live),
or its first store request. Ambiguous same-(ip,persona) concurrent registration is
ignored-late -> both sentinel (never a wrong clean).
- Session cleanup: activity-based TTL sweep (sessions 3600s idle, pendings 120s);
reaping only removes expired entries and never affects another live session.
- account_sync now clears only stale pending for the machine (pre-launch hygiene);
it no longer resets a per-IP mode (there is no per-IP mode any more).
Backend-only: the launcher registration payload (already carries personaId) is
unchanged. Additive; P2 sentinel remains the else-branch and the default.
Tests: matrix A-Q incl. same-IP concurrent (K), same-IP+persona relaunch (L),
same-IP failed-patch (M), late-registration-vs-frozen-sessions (N), TTL expiry (O),
duplicate/idempotent registration (P), and register-before-login pending (Q).
Design + cross-component contract for verified patched-client capability
negotiation: architecture inventory, transport choice (autopatch stdout ->
launcher, sibling /openfut/fifa17/capability endpoint), the versioned capability
and its VERIFIED semantics, autopatch verification states, launcher per-process
state, source-IP binding, the session-stable freeze point, the additive store
switch, the trust model (local preservation, not attestation), the fail-closed
matrix, and P2 retention.
Backend side of the handshake: suppress the synthetic 65534 My-Packs sentinel
ONLY for a session whose client has registered a verified resolver-guard
capability. Additive; the P2 active-sentinel path is retained as the else-branch
and the universal default. Fail-closed everywhere.
- Per-client state keyed by source IP (client_address[0]; the only per-connection
discriminator in this single-account, stateless backend): _FIFA17_STORE[ip] =
{resolver, mode}; mode in {None, "sentinel", "clean-v1"}, guarded by a lock.
- New POST /openfut/fifa17/capability endpoint: accepts only
{"capability":"empty_mypacks_resolver","version":1,...}; unknown capability or
version => 400 and records nothing (=> sentinel).
- account_sync (the launcher's required per-launch call) resets the per-ip record
=> a new FIFA process starts unfrozen with no inherited capability.
- Store topology is frozen at the FIRST /store/purchasegroup per session:
clean-v1 iff a v1 capability is registered, else sentinel; immutable thereafter
(late capability logged + ignored this session; a disappeared capability does
not un-freeze a clean session). This enforces the SESSION-STABLE invariant.
- store_catalog zero-owned-packs branch: clean-v1 emits NO mypacks group (the
client guard routes category -1 to Browse); every other case emits the existing
active 65534 sentinel verbatim. PACK_CATALOG / pack 70 / normal packs / profile
untouched. FIFA-17 only; not lifted into game-independent Core.
- Tests: full matrix A-J incl. concurrency isolation (two IPs, no global leak) and
no cross-process capability leak.
Design: docs/plans/FIFA17_PATCHED_CLIENT_CAPABILITY.md.
autopatch side of the verified patched-client capability handshake: prove, at
runtime, that the empty-My-Packs resolver guard is active for a specific FIFA
process, and advertise it once on stdout for the launcher to relay.
- Add a per-pid guard verification state derived by a pure, testable
guard_state_after(cur_before, orig, patch, wrote_ok, cur_after) returning one
of VERIFIED / UNSUPPORTED_BUILD / WRITE_FAILED / VERIFY_FAILED (NOT_ATTEMPTED
is the pre-evaluation constant). VERIFIED means the live bytes at RVA 0x14858
are 7f 0f (JG) after enforcement (from an applied 75 0f->7f 0f, or already
patched). The existing fail-closed byte guard (guarded_action / STORE_PATCHES_
GUARDED) is unchanged — this only observes the outcome.
- Emit exactly once per FIFA pid: on VERIFIED,
[store-guard] verified capability fifa17.empty_mypacks_resolver=1 fifa_pid=<pid>
otherwise a non-advertising
[store-guard] guard status=<STATE> fifa_pid=<pid> (no capability advertised)
- Capability constants: EMPTY_MYPACKS_RESOLVER_VERSION=1, fully-qualified name
"fifa17.empty_mypacks_resolver".
- Tests: 5 guard-state cases + capability-constant assertions (standalone-runnable).
The capability = "the guard was verified in THIS FIFA process", never merely
"the code is present". Design: docs/plans/FIFA17_PATCHED_CLIENT_CAPABILITY.md.
Land the client-side empty-My-Packs resolver evidence and the session-stability
invariant established by the F3/R1 experiments.
- PART III (F3, CONFOUNDED CRASH): a mid-process sentinel -> no-sentinel flip
left a stale POSITIVE My-Packs ordinal that still took the resolve branch and
crashed at 0x180014882. Preserved verbatim (not a guard failure).
- PART IV (R1, SUCCESS): backend set no-sentinel first, then a FRESH FIFA
process; genuine purchasegroup response ids [1,5,6,7] is byte-identical to the
F3 capture, so client process lifetime is the only changed variable. Store
opens on Browse Packs, no crash, no dialog. Guard PROVEN on the tested build.
- New INVARIANT: empty-My-Packs capability MUST be session-stable -- the server
must not switch a running client between sentinel-present and sentinel-absent
for the My Packs group within one FIFA process, because the client caches the
group ordinal and a stale positive ordinal still crashes the resolver.
- Both no-sentinel captures kept: client_guard (F3) and freshretest (R1).
Backend P2 active-sentinel (65534) remains production default; no capability
handshake is implemented yet.
Port the PROVEN empty-"My Packs" resolver crash-guard into the canonical
autopatch.py /proc-mem patcher. When no `mypacks` purchase group exists, a
fresh FIFA 17 client resolves category id -1; CardsDLL FUN_1800147f0 at RVA
0x14858 (`JNZ 0x14869`, bytes 75 0f) treats every non-zero category as
resolvable, calls FUN_180014420, gets NULL, and dereferences [NULL+0x48] at
0x180014882 (0xC0000005). Rewriting JNZ->JG (7f 0f) preserves positive-category
resolution (EDI>0) while routing zero/negative categories to the existing
Browse/list-all path -> no NULL lookup, no crash, Store opens on Browse Packs.
- STORE_PATCHES_GUARDED table pins RVA 0x180014858 orig 75 0f -> patch 7f 0f.
- Applied every tick, fail-closed via guarded_action(): apply only when the
live bytes are the known original; no-op when already patched; SKIP+log an
unrecognised CardsDLL build (never blindly overwritten).
- Runtime watch loop moved under `if __name__ == "__main__"` so the module
imports cleanly for unit testing; script behavior is unchanged. Existing
ProtoSSL cert-gate and STORE_PATCHES enforcement are byte-identical (indent
only).
- test_autopatch_guard.py: pure test covering PATCH/NOOP/SKIP and pinning the
exact guarded RVA/bytes.
Proven on the tested build (CardsDLL 4706a881...) by a clean fresh-process
no-sentinel A/B (R1). Dormant while the backend active-sentinel is present.
See docs/plans/FIFA17_EMPTY_MYPACKS_CLIENT_FIX.md PART IV.
Investigation + design only (no client/backend/binary changes, no live
Store experiment). Reconfirmed CardsDLL_Win64_retail.dll (4706a881..,
unpacked) against a freshly rebuilt Ghidra project on .105; FIFA17.exe
(29c31cef..) is Denuvo-packed so the Scaleform decision is unreadable.
PART II added to docs/plans/FIFA17_EMPTY_MYPACKS_CLIENT_FIX.md:
- Native category path traced: FUN_18007dab0 (store render, RVA 0x7dab0)
reads screen+0x290; My Packs funnels through FUN_1800147f0 (0x147f0) ->
FUN_180014420 (0x14420, NULL on ordinal miss) -> crash MOV [RDX+0x8] at
0x14882 ([NULL+0x48]), matching the Exp-B minidump. Tab->ordinal map
FUN_180014580 (0=mypacks..5=special); category 0 = list-all (Browse).
- Unopened-pack count is a data-manager singleton (vtbl[0x4d8] get /
[0x4e0] set), reachable from the store resolver.
- Vehicle: existing openfut-hook -> version.dll proxy (already deployed);
reuse ssl_patch signature-scan + connect_hook inline detour. No new loader.
- Preferred strategy A: entry-hook FUN_18007dab0; when the requested
category is My Packs and unopened count==0, force screen+0x290=0 (Browse).
Removes crash + fake 65534 tile + dialog + nav gate; count>0 untouched.
- Ranked B (resolver NULL fallback, higher risk) and C (null-guard, crash-only).
- Build guard: module gate + SHA/PE + signature scan; unknown build -> no
patch, backend sentinel remains fallback.
- First experiment design (needs a later, separately-authorized backend
empty-no-sentinel test mode) + client rollback (config flag / dll swap).
- Keep backend 65534 sentinel deployed until strategy A is verified.
Describes the FIFA 17 client/data model only; not OpenFUT Core assumptions.
Single source of truth for FIFA 17 FUT card families, reconciled against
the authoritative shipped fcc_*.json + staff tables (verified byte-identical
between .105 and this repo, 36/36 sha256).
- docs/CARD_TAXONOMY.md: family -> table/rowcount/subtype/carddbid/cardassetid,
with OBSERVED/INFERRED/HYPOTHESIS/UNKNOWN labels. Corrects four superseded
claims (chem styles are 250-273 not 91-136; 6300/6400xxx are kits not badges;
5004xxx misc and 8010xxx league logos exist). Manager-league precision kept
distinct: shipped table 300-340 (41 rows) vs client enum 300-341 (341 defined,
unshipped). Club-item wire subtype->family mapping preserved as UNKNOWN.
- docs/evidence/fifa17-recon/table-hashes.sha256: 36-file provenance manifest
(31 fcc_*.json + 5 staff tables), combined hash 10f239ad...
Describes the FIFA 17 data/client model only; not OpenFUT Core assumptions.
Full investigation record for bug 6c: baseline + Experiments A/B/C', Candidate F (contradicted), the explicit active-placeholder selection test, the minidump-confirmed CardsDLL crash, and the P2 decision. Marks ROOT CAUSE ESTABLISHED and documents the known UX limitations and the client-side follow-up.
Files: docs/evidence/STORE_TILE_6C.md, docs/evidence/FIFA17_EMPTY_MYPACKS_CLIENT_CONTRACT.md, the four genuine /store/purchasegroup captures (baseline, mypacks70, empty_no_sentinel, active_placeholder), and docs/plans/FIFA17_EMPTY_MYPACKS_CLIENT_FIX.md (client-side design/research).
When the account owns zero unopened packs, store_catalog() emits a synthetic `mypacks` group placeholder (id 65534, absent from PACK_CATALOG). Change its state from "inactive" to "active".
Root cause (bug 6c): FIFA 17's Store/Scaleform path resolves the `mypacks` category even with zero unopened packs (category chosen client-side via the movie's CATEGORY_ID -> screen+0x290; no server field gates it). CardsDLL FUN_1800147f0 then dereferences the resolved group with no null guard, so an absent group crashes the client (CardsDLL+0x14882, [NULL+0x48], minidump-confirmed). An inactive placeholder avoids the crash but makes the Store report the pack unavailable on entry and bounce to the Hub; an active placeholder lets the Store open normally.
65534 stays economy-safe: pack_by_id() returns None, so store_buy()/purchased_items() cannot open it or grant items/coins, and grant_unopened_pack() rejects it. Explicit selection is rejected client-side ("This pack is no longer available") and sends no backend request. This is a FIFA-17 client-compatibility shim (P2), not an EA-authentic representation, confined to the FIFA-17 backend (not OpenFUT Core). A clean zero-pack UX needs a client-side fix (docs/plans/FIFA17_EMPTY_MYPACKS_CLIENT_FIX.md).
Adds regression tests (test_empty_mypacks.py): empty -> one active 65534 placeholder (absent from PACK_CATALOG); non-empty [70] -> no placeholder, genuine pack shown; economy safety; normal packs 1/5/6/7 untouched.
Records the operator-assisted live A/B on the REAL imported club (1949/1962, 13
Legends deferred): real club render, clean pagination, Special filter (1665, 0
base leaked), squad edit persistence + cold relaunch, and rollback->Python->Rust
re-enable. Documents the three real-data fidelity fixes (versioned resourceId
e187cd4, rareflag 626c972, rare=SP filter 6f16a23) and the nation/league/team
model correction (44fcf24). Marks PRODUCTION NOT READY pending .105 Legends name
data for the 13 deferred assets. Aggregate counts only; no private identity.
The club search 'Quality = Special' sends rare=SP, which was a deliberate no-op
('semantics UNKNOWN'), so it returned every card — base golds included. The
rareflag work now grounds it: a special is rareflag > 1 (base rare = 1),
evidence-backed by the FIFA17 taxonomy + the observed profile (base Ronaldo/Messi
rareflag 1; their informs 11/24).
rareflag lives in the FIFA catalog, not Core, so Core cannot filter it:
- map_to_core: rare=SP now sets CoreOwnedQuery.special (host-applied), not
'unsupported'; any OTHER rare value stays unsupported. special is NEVER a Core
/collection param. is_special_rareflag(rf)=rf>1 lives in the adapter.
- handle_club special path: fetch all items matching the OTHER filters (offset/
limit stripped), shape (resolves rareflag), then special_filter_page() keeps
rareflag>1 and paginates the FILTERED set locally (start/count over specials,
not Core's unfiltered page) — no base leakage, no post-pagination drops.
Verified live on the real staged club: rare=SP -> 1665 items (=1949-284 base),
rareflag distribution all >1, zero base leaked; pagination page0==full[0:50],
page1==full[50:100], no overlap. Tests: adapter map rare=SP->special, unknown
rare stays unsupported, is_special_rareflag predicate; host special_filter_page
filter+paginate. adapter 113 + host 25 + importer 25 green; clippy -D clean.
shape_item hardcoded rareflag=1, so all 1949 cards shaped as basic rare gold
regardless of type; informs/specials lost their card art. The dev fixture is
base-only, so this was invisible until the real profile (10 distinct rareflag
values) exposed it on .105.
rareflag is definition-level FIFA identity metadata OBSERVED from the profile
(the raw wire integer, never a guessed marketing label), so it lives in the
FIFA catalog like asset_id/version, not in generic Core:
- ObservedDefinition.rareflag + emitted into the production catalog entry.
- Fifa17CardCatalog RawCard/Fifa17CardIdentity gain rareflag (default 1 when a
base-only catalog omits it, preserving prior wire behaviour).
- Fifa17Identity.rareflag; host resolver populates it from the catalog.
- shape_item emits id.rareflag instead of a hardcoded 1.
Verified on the real staged /club: wire rareflag distribution == source exactly
(0 per-item mismatches across 1949; e.g. rareflag 3 x591, 24 x302, 21 x256).
adapter 111 + host 24 + importer 25 tests green; clippy -D clean. rareflag lives
in the host catalog, not Core, so no re-import was needed.
Evidence (resourceId 169193): its 4 owned copies are IDENTICAL in asset/rating/
position/all attributes and differ ONLY in nation/team/league (and those resolve
inconsistently, e.g. team 240 'Atletico Madrid' under league 16 'Ligue 1'). A
player's club affiliation is an instance-time snapshot, not part of the card
DEFINITION identity.
Correct the model (not a special-case): the definition-consistency gate now
compares a DefIdentity projection (asset_id/version/rating/position/attrs/
rareflag) and EXCLUDES nation/league/team. A club-only difference between copies
of one resourceId is no longer a conflict; a real identity disagreement
(rating/position/attrs/asset) still trips it. The definition's display
nation/league/club use the first-observed copy (deterministic; display-only,
never identity). No --defer-conflict allowlist entry is needed for 169193 now.
On the real profile: conflicts 1->0, 169193 reclassified conflict->NoName
(still deferred, unnameable), supported still 1681, deferred instances still 13,
BLOCKERS none without any --defer-conflict flag. 2 new tests (club-only diff is
not a conflict; rating diff still is). crate suite 25 green; clippy -D clean.
shape_item emitted resourceId/definitionId = asset_id (base), collapsing every
versioned (special) card onto its base definition on the /club and squad wire.
The dev 32-card fixture is base-only (version 0), so Slice 7 never exposed it;
the real profile (1531 versioned cards) did.
Fifa17Identity now carries resource_id (= (version<<24)|asset_id, == asset_id
for a base card). shape_item emits resourceId/definitionId from resource_id and
assetId/cardassetid from asset_id — versioned and base stay distinct. The host
resolver populates resource_id from the catalog's reconstructed resource_id
(the catalog already parsed version; it was dropped before shaping).
Regression test: versioned 117617092 (v7 of asset 176580) shapes resourceId/
definitionId=117617092, assetId/cardassetid=176580.
Verified on the real staged /club: 1949 items, wire-id set exact, 0
wire->resourceId mismatches, 0 duplicate-multiplicity mismatches vs the source
manifest. adapter 111 + host 24 tests green; clippy -D warnings clean.
FIFA17-specific orchestration that installs the real profile across BOTH durable
stores (openfut-identity + Core SQLite) recoverably and idempotently, handing
Core only a GENERIC request (all FIFA17 semantics stay in this adapter layer).
apply module:
- owned_item_id(persona, wire) = deterministic UUIDv5 from a private namespace;
identical in BOTH stores (identity core_id AND Core owned_cards.id), so the
running host's wire->owned reverse lookup resolves exactly what Core stored.
- plan_apply(report, raw_profile, fp): pure translation to a GenericImportRequest
(card_id = fifa17_<resourceId>) + the preserved (owned_item_id <-> source wire)
mappings + watermark. Canonical squad + Fifa17SquadExtensionV1 are built by the
SAME adapter code (parse_squad_put + build_squad_write) the retail-validated
live squad path uses. Refuses if the report has blockers.
- Two-store protocol (apply): staging gate -> local Core-preflight mirror ->
identity dry-preflight -> idempotent seed (insert_existing_mapping per instance
+ set_watermark) -> ONE generic Core import transaction (spawned binary) ->
cross-store post-validation -> completion record. A crash after identity
seeding re-converges on re-run (idempotent mappings + Core already_imported):
no cleanup, no reminting.
- gate_staging: deferred players are ABSENT from an import; allowed only for a
staged run behind --allow-deferred-players-for-staging (never a silent default;
prints an INCOMPLETE banner). Production requires zero deferred instances.
CLI: --apply (with --emit-content, --core-bin, --core-db, --core-data,
--identity-store, --allow-deferred-players-for-staging). Depends on
openfut-adapter-fifa17 + openfut-identity + uuid(v5).
Proven end-to-end on the real 33068179/CAGE profile (staged): first apply
imports 1949 supported instances (293 base + versioned), 11/11 f433 squad +
opaque extension, coins 28,112,944, fingerprint 8dc5582d2414af28; re-run is an
idempotent no-op (already_imported, DB unchanged); staging-not-default refuses
before any write; every OwnedItemId is an opaque UUID; identity wire-id set ==
source supported set exactly (0 minted, 0 dropped); 13 deferred instances leak 0.
10 new apply tests (determinism, request/mapping/squad translation, blocker
refusal, staging gate both ways, local preflight, identity seed/dry/postvalidate/
idempotency, conflict detection, graceful spawn failure). clippy -D warnings
clean; crate suite 23 tests green.
Fold entity resolution (nation/league/club id->name via committed tables,
mirroring seed_fifa17_cards.py) and quality-tier rarity into the analysis, so
the supported set is honest about unresolved entities too. Add an explicit
--defer-conflict <rid> allowlist: a reviewed conflict (169193) defers, any NEW
conflict still hard-fails (defer never becomes a silent conflict suppressor).
--emit-content writes three files, PUBLIC content separated from PRIVATE account
state: fifa17-production-cards.json (Core CardDefinition[] keyed fifa17_<resourceId>,
base+versioned, tier rarity, profile-derived, no promo labels), a versioned host
identity catalog {card_id:{asset_id,version}}, and a private import manifest
(supported instances' wire ids + deferred set with reasons + preserved watermark
+ target profile + snapshot fingerprint). Emit refuses while blockers exist.
Real profile (33068179/CAGE): 1681 supported defs (150 base + 1531 versioned),
9 NoName deferred, 1 approved-deferred conflict (169193, 4 copies), 1949
importable instances, watermark 100004617 -> next 100004617, active squad f433
11/11 supported. fmt + clippy -D warnings clean; 13 tests.
openfut-identity:
- insert_existing_mapping(game,kind,core_id,external_id): preserve an existing
external wire id instead of minting; idempotent for an identical mapping,
rejects conflicting forward/reverse with IdError::Conflict, persists atomically.
- persisted per-scope allocator watermark (set_watermark/watermark_for) so a
future mint continues past the source high-water even across burned-id gaps;
next id = max(base_floor, live_max+1, watermark). Backward-compatible on-disk
format (legacy bare [Row] still loads). +4 tests (10 total).
openfut-import-fifa17 (new): read-only dry-run analysis of a real FIFA17 Python
profile for a faithful Core import. Enforces disjoint item-class balance;
proposes profile-derived CardDefinitions keyed fifa17_<resourceId> (base vs
versioned never collapse) with a resourceId-group consistency gate (hard-fail on
disagreement, never pick a winner) and honest buildability (roster name +
version formula + metadata, never fabricated); plans owned-instance identity
(preserve Python wire ids, preserve nextItemId watermark); checks active-squad
coverage. --apply/--emit-content refuse to write in this phase. 11 tests.
Real profile (33068179/CAGE) dry-run: 1982 items balance (1962 players + 17
consumables + 3 staff); 1681 supported defs (155 base + 1535 versioned), 9
NoName unsupported, 1 hard conflict (resourceId 169193: one of 4 copies has a
divergent nation/team/league); 1949 importable player instances, watermark
100004617 -> first new alloc 100004617; active squad f433 fully supported.
fmt + clippy -D warnings clean.
Record the staged retail A/B: FIFA consumed the Rust squad path end-to-end
(userMassInfo overlay, in-game swap -> squad-replace {"id":0}, formation
f442->f433 persisted to Core, cold relaunch returned the persisted squad),
with Python rollback / Rust re-enable proven by host-log presence. Note the
OPENFUT_DEV_CONTENT_GAMES=fifa17 startup requirement (silent empty /collection
if omitted) as a needed deployment/preflight assertion.
"TLS HANDSHAKE FAILED: ... unexpected EOF" is exactly how the certificate
mismatch presented -- the defect that cost three live gate attempts and was
invisible everywhere else. It is the one line this project has learned to
treat as serious.
A reachability probe forges it for free: TcpStream::connect followed by a
drop opens the connection and closes without sending a byte, and the
acceptor reports that as "unexpected EOF". The launcher's preflight makes
two such probes per run. On 2026-08-12 they produced ten of these lines and
sent a whole session diagnosing a client-side fault that did not exist --
autopatch, ptrace_scope and client_arm.sh were all investigated before the
pairing of the timestamps gave it away.
Classify before the acceptor sees the connection: peek one byte, and treat
EOF-before-any-byte as a probe with its own quiet line. A timeout is
deliberately NOT a probe -- a slow or broken client must still reach the
acceptor and produce a real diagnostic, since misclassifying a fault as
benign would defeat the point.
The counter exists because the test needs it. A probe produces no response
either way, so a test written against client-visible behaviour passes with
the classification deleted; asserting on a count is what makes the mutation
detectable. Verified: all three mutations (drop the classification, treat
undetermined as a probe, treat a speaking client as a probe) are killed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Trailing rustfmt reflow of two catalog unit-test assertions (no logic change); clears the adapter dirty state so verify-build-identity.sh passes for the UTAS A/B.
Migrate the active-squad READ off the Python oracle to the existing Core-backed projector, completing the squad authority (read + write + /squad/list + userMassInfo overlay) on one projector.
- classify: GET /ut/game/<t>/squad/active -> Route::SquadActive. Numeric GET /squad/<n> stays on Python (no Core multi-squad model yet).
- handle_squad_active returns the projector object via user_mass_info_squad(v, persona) — byte-identical to userMassInfo.squad; degrades to an empty overlay on stale/missing/Core-error, never falls back to Python.
- persona: new REQUIRED OPENFUT_PERSONA_ID (non-zero) on HostConfig, injected not baked, must match LSX/Blaze/POW/UTAS identity.
- tests: squad_active parity test; classify updated; README config table + A/B command.
fmt + clippy -D warnings + tests (24 host + adapter) green.
Formatting reflowed two mutation target lines onto multiple lines; update
the battery anchors (adapter #14 kit lookup, host #19 Content-Length push)
to the new unique substrings. Both batteries kill all mutants again
(adapter 14/14, host 20/20).
Formatting-only. Runs the project formatter over the files authored/edited
this session (host lib+tests, adapter item/squad_ext/squad_projection/mod +
projection test). The intentionally-preserved dirty catalog.rs and
pre-existing /club-era host drift beyond these files are out of scope.
The capture sanitizer redacted tokens/device ids but left the lab Host
address (10.10.0.x) in three committed UTAS fixtures, tripping
scripts/check-no-lab-addresses.sh. Replace with an RFC 5737 TEST-NET
address (192.0.2.120) per the guard's own doctrine. Host header is
plaintext metadata only (tests decode body_b64), so no test is affected.
Pre-existing leak (fixtures committed at 0b66662/eb8311a); no lab address
was introduced this session.
mutation-battery.sh injects each of the 20 required wrong behaviours into
the committed host (or the adapter it composes) source, runs the one host
test that must catch it, and requires a non-zero exit (killed), reverting
via git after each. Adds a duplicate-definition host round-trip test.
Kills: authz-skipped, authz-loop-empty, failure-masked-as-success,
PUT-as-partial-diff, extension-dropped, stale-accepted, missing-fabricated,
overlay-clobbers-{userInfo,settings,pile}, list-separate-shaping,
captain-as-resourceId, kit-by-slot, client-eval-dropped, host-trusts-fp,
per-slot-N+1, squad/active-to-Rust, duplicate-collapse, stale-content-length,
python-squad-after-failure. 20/20 killed.
Extend the UTAS migration host to own the squad slice, reusing the exact
production identity path (Fifa17CardCatalog + ExternalIdentityStore) that
/club uses, so /club, /squad/list, userMassInfo and PUT all agree on
wire<->owned identity.
Routing (classified ONCE, no try-Rust-then-Python):
PUT …/squad/<n> -> Rust (numeric id; …/squad/active stays Python)
GET …/squad/list -> Rust
GET …/userMassInfo-> Python proxy, ONLY .squad overlaid
everything else -> Python verbatim
CoreAccess gains read_squad_ext / replace_squad / all_owned (HTTP to the new
Core /squad/ext + /squad/replace routes).
PUT pipeline: parse -> build_squad_write (reverse-resolve every wire id;
refuse unresolved/duplicate) -> AUTHORIZE every resolved owned item against
the active club (identity resolution is NOT authorization; a valid wire id
owned by another profile is rejected before any mutation) -> Core atomic
replace+extension -> exactly {"id":0}. No Python fallback on failure; no
host-side second extension store; no fingerprint recomputation.
Read path assembles ONE projection input (one read_squad_ext + one batch
all_owned; no per-slot lookup) and runs the single adapter projector. Both
/squad/list and userMassInfo.squad derive from it. Fresh projects; Stale is
never applied; Missing is never fabricated — both are prominent integrity
failures, never served from Python (no split authority). The overlay
replaces only .squad and preserves userInfo/settings/userData/
pileSizeClientData, fixing Content-Length.
Tests: routing, full-replacement + exact ack, unknown/foreign/duplicate item
rejection with Core unchanged, idempotent repeat PUT, coupled read-after-
write (list + userMassInfo agree, real resourceIds + stable wire ids),
overlay field preservation, bounded no-N+1 reads, Stale/Missing integrity.
Exposes GET /squad/ext and PUT /squad/replace so the UTAS host can read
and atomically persist the FIFA17 squad canonical+extension state over
HTTP. Core commit sits atop the frozen contract 615c5fd (branch
rust-migration/squad-ext-routes); no domain change. The preserved dirty
openfut-core worktree (eab522a) is intentionally left untouched, so root
status still shows 'M openfut-core' as before.
mutation-battery.sh injects each of the 14 required wrong behaviours into
the committed source, runs the one invariant test that must catch it, and
requires a non-zero exit (mutant killed), reverting via git after each.
Kills: kit-by-slot, captain-as-resourceId, chemistry-reconciled,
custom-regenerated, index-derived, stale-accepted, missing-fabricated,
faked-asset-id, duplicate-instance-collapse, PUT-as-slot-diff,
wire-id-in-canonical, projector-bypasses-shared-shaper, schema-version-
ignored, player-state-keyed-by-definition. 14/14 killed.
fut::squad_projection is the ONE projector for every squad read shape.
project_squad(canonical squad + Fresh extension + owned items) -> the FIFA
17 squad wire object; user_mass_info_squad and squad_list are envelope-only
wrappers over the same output (no per-endpoint domain model).
Design guarantees exercised by tests:
- purity / no N+1: consumes a host-assembled input (read_squad_with_ext +
one batch owned-cards fetch + in-memory card defs); no per-slot lookup
- shared shaper: every occupied slot is shaped by fut::item::shape_item,
so squad items and /club items cannot drift
- Fresh -> full projection; Stale -> never applied (verdict surfaced);
Missing -> explicit, never fabricated
- captain projects as the resolved WIRE id (never resourceId); index and
formation round-trip verbatim; kit follows the player; two owned copies
of one definition stay distinct
Adds committed sanitized fixtures decoded from the squad session capture
(swap, f433, persisted userMassInfo.squad read, squad/list) and
tests/squad_projection.rs: baseline / swap / formation-change / persisted
read-after-write round-trips asserted by ownership class (canonical,
extension, shadow, derived identity), plus one-projector no-divergence.
Add fut::squad_ext::Fifa17SquadExtensionV1 — the versioned, adapter-owned
payload Core stores opaquely alongside the canonical squad. Carries the
FIFA-only wire state that is not Core-canonical:
- custom[] opaque 33-int string, round-tripped verbatim
- squad_type observed FIFA token
- kit_numbers keyed by owned_card_id (kit follows the PLAYER, proven
by the swap/formation captures), never by slot/definition
- manager opaque item ref (not a squad player; not shaped)
- kicktakers opaque role refs; relationship to captain UNKNOWN, so
preserved verbatim and never normalized to the captain
- client_reported chemistry/rating/starRating shadow, never authoritative
from_payload enforces the payload schema version first (distinct from Core's
DB schema); an unknown version is rejected, never coerced.
build_squad_write turns a parsed PUT + host wire->owned resolver into a
canonical ProposedSquad + extension, refusing on unresolved ids or a
duplicate owned item. Identity resolution is explicitly NOT authorization.
Refactor the 550a59d parser scaffold: ProposedSquad is now pure canonical
(FIFA-only + shadow fields moved to the extension); the canonical formation
is the FIFA wire token verbatim (drop the lossy f442->"4-4-2" map that
could not even represent f433) so formation and index round-trip exactly
with no derivation. Bench split is the fixed 23-slot array convention.
Move the per-item card shaper (CoreOwnedItem, Fifa17Identity,
ItemIdentityResolver, ShapeStats, shape_item) out of club_response into
fut::item so /club and the upcoming squad projection emit byte-identical
items from one source of truth. club_response keeps only the /club
{itemData:[...]} envelope and re-exports the moved types for API
stability. shape_item is now pub; no behavior change (all /club and
oracle-parity tests unchanged and green).
Adds item-shaper tests: full-field identity mapping and the duplicate
owned-copy invariant (two instances of one definition keep distinct wire
ids, share one asset id).
scripts/seed_fifa17_cards.py: deterministic pipeline from committed FIFA17 data
(pool.json + roster.json + leagues/nations/teams tables) -> the card-definition
identity catalog. CardDefinitionId is opaque + deterministic (fifa17_<asset>),
version 0 (base cards only; resource_id == asset_id). --check mode diffs against
committed output (drift-detection mutation-proven). Provenance embedded.
Generated openfut-adapter-fifa17/data/fifa17-card-identities.json: all 17,563
base assets. Semantic definition coverage (to /tmp, not committed here): 17,547
resolvable; 16 skipped for missing roster name (reported, never fabricated).
Adapter loads the committed catalog (test: 17,563 entries, Ronaldo fifa17_20801
-> asset 20801 v0). Phase commit 3/5. NOT owned inventory: this is 'which cards
exist', not 'which the user owns'. Core content seeding + dev-owned set next.
fut::catalog — Fifa17CardCatalog maps a semantic CardDefinitionId to a FIFA 17
render identity (resource_id = (version<<24)|asset_id; version 0 => resource==
asset). Versioned JSON (schema_version=1, game=fifa17); validates schema/game,
rejects asset_id > 24 bits, and rejects two card ids claiming one resource_id.
Unknown definitions resolve to None (callers drop, never fabricate).
Fifa17WireItemIdPolicy carries the owned-item namespace (base 100_000_000,
first id 100_000_001, per the oracle) supplied to the generic store.
Adds serde derive to the adapter. 8 catalog tests; 3/3 mutations killed
(resourceId-drops-version, conflict-detection-off, asset-range-off).
Phase commit 2/5. No card->asset DATA shipped: the synthetic Core catalogue is
unmappable (see seed plan); the loader + format land now, population later.
openfut-identity: durable, reversible (game_id, entity_kind, core_id) <->
external wire id mapping. Game-independent infrastructure (adapters supply the
numeric policy via base_floor; the store guarantees stable/unique/reversible/
game-scoped/persistent/atomic/explicit). JSON-file backed behind an
ExternalIdentityStore trait (SQLite can drop in later); parking_lot-guarded,
atomic temp+rename persist, rejects a torn reverse-duplicate on open.
Core never learns FIFA integers; only the host/adapter that owns a game
boundary uses this. 6 tests, 4/4 mutations killed (same-id-for-two-items,
lost-on-restart, broken-reverse, dropped-game-scope). Phase commit 1/5.
openfut-utas-host: the first live UTAS host. Serves GET /ut/game/<title>/club
from OpenFUT Core via the FIFA17 adapter and reverse-proxies every other UTAS
route verbatim to the Python oracle. Plaintext HTTP/1.1 keep-alive (no TLS);
route classification before execution; a Core error on /club degrades to an
empty page and never falls back to Python. CoreAccess is a host-owned boundary
(the adapter stays transport-agnostic).
openfut-adapter-fifa17::fut: owned_query (wire parse + FIFA id->name mapping,
unknown id = hard error), entities (id<->name from committed tables), and
club_response (FIFA _item shaping; drops items lacking a real FIFA asset id,
never fabricates one).
openfut-core submodule advanced to the reconciled trunk (6acae54 = 8c8a4116
multi-game + eab522a replace_squad/SquadRules + the /club semantic query).
11 host tests + adapter fut tests; 10/10 host mutations killed. rare=SP UNKNOWN.
Retail rendering of Core inventory still blocked on the Core-card->asset-id
identity decision (next phase).
Controlled retail capture, one criterion at a time, cleared between each.
47 transactions. Every filter the My Squad picker sends is now known from
the wire rather than guessed.
Route: GET /ut/game/fifa17/club -- the picker hits UTAS and reuses the
general club-inventory route.
level=any|gold quality lowercase, ALWAYS present
rare=SP "Special" uppercase, OMITTED when off
position=ST position uppercase, omitted when off
nation=52 entity id numeric
league=13 entity id numeric
team=5 entity id numeric, NESTED under league
sort=desc client constant; the UI has no sort control
start=/count=11 pagination
Two encoding families: short string enums, and numeric FIFA ids. The ids
must never reach Core. Filters compose as plain ANDs in one query --
string and id filters alike -- so each maps independently.
THE ROOT CAUSE IS SELF-AMPLIFYING.
club_route honours type, team and league; it never reads start, count,
level, sort or year. Because start is ignored, every page returns the
same full set, so the client concludes the page was full and asks for the
next one. One scroll produced 22 requests and 6.2 MB, stopping at
start=200 only because the client gave up -- against a filtered set of 32
items that should have been three pages.
That also explains why the bug reads as erratic rather than broken:
league=13&position=ST returns every Premier League player instead of
Premier League strikers. Plausible, wrongly sized, hard to notice.
Measured filtered sets, from the real cluttered club -- these are the
acceptance test for the fix:
unfiltered 1962
league=13 350
league=13&team=5 32
Two client behaviours worth carrying forward: the picker fires a query
per highlighted entry, not per selection (two requests for one club
pick), and parameter ORDER is not stable, so parsing must be key-value.
FIXTURE SIZE: bodies over 4 KB are truncated in the committed fixture,
with body_full_len and body_full_sha256 retained, because the same 1.1 MB
club response repeats ~25 times and its hash already proves identity.
13.3 MB -> 247 KB. The raw .ofcap keeps every byte, privately and
gitignored. Truncation is recorded per transaction so a trimmed fixture
is never mistaken for a whole response.
Audited across all three identifier surfaces -- headers, JSON bodies,
query strings -- before and after the size change: no leaks. 6/6
sanitiser mutations still killed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
24 transactions across 11 connections from a retail session: login,
hub, one pack open, two squad saves, a quick-sell, with before/after
state manifests. Raw .ofcap stays gitignored at 0600; the sanitized
corpus is committed as adapter fixtures.
TWO GAPS FOUND BY AUDITING THE OUTPUT, NOT BY TRUSTING THE SANITISER.
1. `POST /ut/auth` carries `macAddress` and `deviceId`. Session tokens
were being redacted correctly and these were not. A committed fixture
is a published fixture.
2. Then, with those fixed, the audit fired AGAIN on the file about to be
committed: `GET .../phishing/trusteddevice?deviceId=...` puts the id in
the QUERY STRING. Three input surfaces carry identifiers -- headers,
JSON bodies, and query strings -- and the sanitiser knew about two.
Both fixed in the tool rather than by editing the file, with a
regression test and a mutation for the query path.
AND A THIRD ARTEFACT MIX-UP, in the mutation harness itself. It reported
the query-redaction mutation as SURVIVED while a hand-run of the same
mutation killed it. Cause: the harness pointed at a stale scratchpad copy
of the test that pre-dated the query assertion, so it was faithfully
testing the mutated tool against a test that could not detect the
mutation. That is the same class as the build guard checking the wrong
binary and cargo reusing a binary compiled from mutated source -- the
third instance today of measuring the wrong artifact. The harness now
resolves ROOT from its own location and runs the COMMITTED test; the
stale copy is deleted.
Harness committed as scripts/mutate-utas-observe.py so this is repeatable
rather than a thing that happened once in a scratch directory. 6/6 killed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
UTAS needs a real request/response corpus before any Rust is written: it
is where protocol shape and FUT state start being coupled, so guessing is
worse here than it was for Blaze. The oracle truncates logged bodies at
~200 chars, and raising that cap would mean editing the behavioural
specification to make it easier to copy -- backwards. A proxy gets the
same evidence and leaves the oracle untouched.
THE DESIGN RULE: TEE, DO NOT REBUILD.
UTAS is plaintext HTTP/1.1 on ThreadingHTTPServer, so keep-alive,
pipelining and chunked transfer are all live. A proxy that parses a
request and re-emits it can corrupt the traffic it exists to observe --
and that corruption would present as a UTAS bug, pointing the
investigation in exactly the wrong direction. So bytes are copied
verbatim in both directions and a second copy goes to disk; transactions
are reconstructed later, offline, from that copy. A parser bug therefore
spoils the record and never the session.
Standalone, NOT in the container, so the same tool can later sit in front
of a Rust UTAS host and replay an identical captured request against both.
Two layers, as with the Blaze captures: raw/*.ofcap is exact bytes at mode
0600 and gitignored; sanitized/transactions.jsonl is the committed
artefact. Bodies are preserved EXACTLY and sanitised second -- only
known-secret headers and JSON keys are replaced, structure is never
reshaped, and every redaction is recorded in the transaction so a reader
knows what was touched.
Captured per transaction: connection id, sequence, relative and wall
time, elapsed ms, method, path, query, HTTP version, headers IN RECEIVED
ORDER as pairs (a dict would drop duplicates and ordering), raw body and
length for both directions, status, and observed keep-alive.
Verified as two independent properties, because they fail differently:
transparency (bytes through the proxy identical to bytes direct, Date
masked, with the mask asserted to have fired) and fidelity (parsed
transactions match what was sent, including a dechunked response and a
300-byte POST body). 5/5 mutations killed, including "record but do not
forward", "drop the last byte of every chunk" and "stop redacting".
scripts/test-utas-observe.py is committed alongside it: a capture tool
nobody can re-verify is not evidence.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Queued cleanup, run only AFTER the roster gate closed in both directions,
so the live A/B changed exactly one thing.
The two `drain_body` implementations were character-for-character
identical, so the extraction is a move. What it guards is not cosmetic:
answering while the client is still sending leaves unread data in the
receive queue and Linux turns the close into an RST rather than a FIN --
invisible in any comparison of the response, and worth two live gate
attempts to find. Behaviour that must be identical across hosts gets one
implementation, the same reasoning that produced openfut-tls.
SCOPE IS DELIBERATELY NARROW. Only the byte-identical part moved. The two
head-reading loops are NOT identical and stay where they are:
redirector roster
head cap 65536 16384
read chunk 4096 1024
on error abort proceed if any bytes arrived
Those differences are probably accidental, but each host is gate-proven
with the values it has. Unifying them would be a behaviour change wearing
a refactor's clothes -- exactly the mistake this project has already paid
for. They converge later as their own change with their own gate, or not
at all.
Purity shown, not asserted: every existing test in both hosts still
passes (426 workspace tests), and 7/7 mutations are killed, including
three in the SHARED crate that must break both hosts at once and one per
host that skips the drain call.
Three test cases neither host had now exist, because the extracted code
finally had somewhere to be tested directly: a malformed Content-Length,
an unterminated head, and a lookalike header. That last one matters --
`X-Original-Content-Length: 99` would drain 99 bytes that were never sent
if the match were `contains` rather than `starts_with`, and a mutation
confirms the test catches it.
Also fixes a race this run exposed in openfut-tls's own tests: keypair()
returned early if the certificate file existed, but wrote the certificate
BEFORE the key, so a parallel test could observe a cert whose key had not
landed. It failed one run and passed the next -- the kind of flake that
gets rerun instead of fixed. Now generated once per process via OnceLock,
key written first, and the suite was repeated five times to confirm.
Nothing deployed and nothing restarted: the running redirector and roster
are still the gate-proven binaries.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
redirector.sh and the coming roster.sh needed the same five rules, each
of which cost something to learn:
* resolve /proc/PID/exe; never match a command line. `pkill -f` /
`pgrep -f` match any shell whose ARGUMENTS mention the name, including
the shell running the command. That has killed this session's own
shell twice, and is now banned in migration tooling -- the helper
contains no `-f` matching and the header says why.
* `readlink`, not `readlink -f`. After a rebuild the link reads
"<path> (deleted)" and -f resolves it to nothing, so the orphan check
goes blind to exactly the long-lived processes it exists to find. Two
orphans hid there, one serving the wrong certificate.
* stop PROVES the process is gone and the port free.
* an ambiguous binary is an error for start/verify but NOT for
stop/status: rollback must never be blocked by a question about the
build tree.
* verify the RUNNING process's commit, not the artifact on disk, which
a rebuild can silently advance past.
Copying those into a second script would have been the same mistake as
copying the TLS setup. Instead scripts/host-lifecycle.sh owns them and a
service supplies four facts: name, crate, executable, port variable.
redirector.sh goes from 178 lines to 26 and roster.sh is 24, with no
behaviour change -- the refactored redirector.sh still sees the live
armed process (pid 830736, port 42227) and still refuses correctly
because HEAD has moved past it.
Paths are unchanged (rundir, pidfile, portfile, commit stamp, log), so
the currently running redirector stays manageable across this refactor.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Second consumer of openfut-tls, and the reason it was extracted first.
This host contains no roster content and no cipher choice: the adapter
owns the 67 bytes and the observed TLS profile, openfut-tls owns the
acceptor, and this crate owns accept/read/drain/write/close.
Lifecycle was MEASURED, not inherited. The obvious mistake here would
have been copying the redirector's 300ms dwell because the other host has
one. A probe against the oracle says otherwise:
dwell after responding 0 ms (redirector: 300 ms)
request body drained POST answered only once it arrives
close clean FIN, never RST
keep-alive none one request per connection
The probe ran against a REPLICA of roster_server.py loaded from its own
source, not against :8081 -- http.server.HTTPServer is single-threaded
and FIFA was mid-session, so holding a connection open to measure the
close would have stalled the game's poll and could have surfaced as the
squad-update error. The replica was then confirmed byte-identical to the
live oracle under masking, the 1-byte delta being the container's Python
version in the Server header.
Differential against the live oracle, every field identical, with the
Server header compared UNMASKED:
GET 230B HEAD 163B POST 163B
drained=True reset=False answered_before_body=False
keepalive: second request accepted by the socket, never answered
Testing follows the redirector's hard-won rule: where a property is
visible both to the client and inside the host, it is asserted inside the
host via ConnOutcome. A client-side check cannot tell "drained" from "not
drained" -- it reads the buffered response either way -- and that exact
mistake let a mutation survive once already.
9 parity tests, 6 unit tests, 5/5 mutations killed, including "answer
before draining", "hold the connection open like the redirector" and
"inherit the redirector's 300ms default".
drain_body is duplicated from the redirector deliberately. Unifying it
means editing the redirector, and the roster A/B must change exactly one
thing. Extraction is scheduled for after the roster gate closes.
Not deployed and not switched: Python still serves :8081.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The redirector was the only host that spoke TLS, so its TLS lived inside
it. The roster host needs the same listener, and that made the choice
explicit: share this code or copy it.
Copying it is what already went wrong. On 2026-08-11 the Rust redirector
served one certificate while the container served another. ProtoSSL
caches the server certificate per backend, so the redirector -- the first
TLS connection of a session -- decided what the client expected, and
every later service failed its handshake. Silently: Python's socketserver
swallows ssl.SSLError as OSError. Three gates went to it. One place to
configure TLS is the structural fix, so it exists before the second host
does rather than after.
Split along the line the architecture already draws:
openfut-tls how to build an acceptor. Game-independent.
Knows nothing about which suites any client
offers.
adapter-fifa17::tls what FIFA 17 was OBSERVED to offer: the six
enabled suites, the two refused, the TLS 1.2
window, the EA SNI. Plain strings, so the
adapter keeps its lean dependencies -- reading
a card table should not build OpenSSL.
redirector-host joins the two. Chooses no cipher of its own.
Behaviour is unchanged, and shown to be:
* tests/fifa17_tls_profile.rs carries over every case from the deleted
module -- FIFA's eight suites negotiate AES256-GCM-SHA384, each enabled
suite works alone, RC4-only is refused, ECDHE-only is refused. Deleting
a module must not quietly delete its evidence.
* one test pins the composed values literally against the host as it was
when gates 1-14 passed. A "pure refactor" that cannot fail is not a
claim, it is an assumption.
* the rebuilt binary self-tests to the same TLSv1.2 / AES256-GCM-SHA384
the retail client negotiated at 17:09 today.
Two improvements fall out of having one place to look:
* the startup banner now prints cert_sha256. The mismatch above raised no
error at startup and broke the client much later with nothing logged;
it is now the first line of the log.
* tls_min/tls_max print as TLSv1.2 rather than SslVersion(771). This line
is gate evidence and gets read by people.
Nothing deployed and nothing restarted: FIFA is mid-session on the
running redirector, which is untouched.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The old check was `grep easw /etc/hosts && echo ok`. It passed on ANY
matching line -- including a line that shadows ours. glibc returns the
first match, and the sed above only deletes lines this script wrote
(`# openfut`), so a foreign entry earlier in the file wins forever and
re-running the script never helps.
Observed today: a leftover `127.0.0.1 easw.easports.com` from the
single-machine era, before the backend moved to its own host. Every arm
reported "/etc/hosts ok" while the name resolved to loopback.
Now it resolves the name -- the same call the game makes -- and compares
address to address, so a server given as a hostname is handled too. On a
mismatch it prints the offending lines with line numbers and says how to
fix them.
It does NOT delete them. This script writes one tagged line and owns only
that line; silently removing entries a user put there by hand is a bigger
hazard than the shadowing it would cure.
Reported as a warning, not an error, because it is survivable: the
responders advertise the server address, so the game stops using this
hostname after the first redirected contact. FIFA reached the FUT hub
today with this exact misconfiguration in place. Claiming it is fatal
would be wrong, and a check that overstates its findings gets ignored.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Two defects, both found by the guards misfiring rather than by reading:
1. BIN preferred target/debug and fell back to release only when debug was
absent. `cargo build --release` therefore produced a correct binary while
the script kept inspecting a stale debug one, and the build guard refused
with a message naming a commit nobody was trying to run. The guard was
right that something was stale — it just pointed at the wrong artifact.
Disagreement between the two is now an explicit refusal naming both, with
OPENFUT_REDIRECTOR_BIN as the deliberate override.
The refusal is recorded at load and raised only by `verify` and `start`.
`stop` and `status` must work in any build-tree state: rollback can never
be blocked by a question about which artifact would have been started.
2. `verify-running` read the stamp file without checking the process still
existed. The stamp outlives the process, so after a stop it reported on a
corpse — either "identity OK" or a REFUSAL naming a commit, both implying
something was running when nothing was.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
`cmd_status` has two paths. The `--name` path filters on an exact tag and
works. The no-name path — the "show me every switch on this box" survey,
which is how an orphan switch under a different name would be found —
built its python with shell quote-juggling and never closed the string
literal, so it died with a SyntaxError every time.
It failed loudly (rc=1, a traceback) rather than reporting "no rules", so
it never lied about the state. But it also meant the survey path had
never once run, which is the more useful lesson: every branch of a safety
tool needs exercising, not just the branch the happy path takes.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
`list_procs` matched on `readlink -f /proc/PID/exe`. Once the binary is
rebuilt -- which happens constantly here, `cargo test` alone is enough -- the
link reads "<path> (deleted)" and -f resolves it to something that matches
nothing. The scan then finds zero processes, so `start`'s orphan check passes
and a second instance can be launched alongside a stray.
Not theoretical. Two orphans were running undetected tonight:
pid 592731 :42327 a stale-cert redirector left from testing check-tls-parity,
still serving F9:16:1A -- the exact certificate whose
mismatch cost three live gates
pid 542693 :42230 a Blaze sidecar debug build from 03:12
Neither was in a client path, so neither was doing harm, but a stray listener
serving the known-bad certificate is precisely what should never sit around
unnoticed.
Fixed by using plain readlink and stripping the " (deleted)" suffix. Shown both
ways: with the bug `status` reports no processes at all for a live pid; with the
fix it reports 604454. The pidfile path was unaffected, which is why `stop` kept
working and hid this.
It is a TOTAL -- every packet reaching the chain counts there, including ones
already counted by a named-port rule. The old wording invited reading the number
as a remainder, which is how 15 unexplained attempts got misread earlier.
Next component in the migration order (Roster -> LSX -> UTAS). Adapter layer
only: no host, no runtime replacement, nothing armed.
The response is shaped as much by http.server.BaseHTTPRequestHandler as by the
oracle's handler code, so it is captured over the wire rather than reasoned
about:
* HTTP/1.0 status line -- protocol_version is left at its default, so the
reply is 1.0 even though the client asks for 1.1
* send_response injects Server: and Date: BEFORE the handler's own headers
* POST answers with headers only: the handler writes the body `if method ==
"GET"`, so a POST advertises Content-Length: 67 and then sends nothing
That last one is preserved, not corrected. It looks like a bug, but "obviously a
bug" has been the wrong call before in this port, and a test now asserts it so a
future cleanup has to argue with something.
Date and Server are volatile and are MASKED in the fixture rather than dropped,
so their presence and position are still asserted. Server is additionally
recorded verbatim: it carries the container's Python version, so a drift away
from roster::ORACLE_SERVER fails a test instead of silently changing every byte
we emit.
generate_roster.py --check FAILS when it cannot reach the oracle rather than
passing, and mutation-testing the mutation harness itself caught two "surviving"
mutations that were really sed no-ops. With application verified, all four
mutations (header order, Content-Length, XML body, Connection) are killed.
Three times now the same sequence has broken the client path: a build guard
correctly refuses to start the Rust replacement, and the `switch on` that
follows in the same script arms anyway, because it never checked whether
anything was listening. The redirect then lands on a closed socket and the
working Python service is bypassed for no benefit.
`on` now refuses unless the target port is listening. ALLOW_DEAD_TARGET=1
overrides it for arming ahead of a service that is about to start, but that has
to be deliberate. Verified both ways: rc=2 and nothing installed against a dead
port, rc=0 and two rules with the override.
The watchdog now writes a pidfile. Stopping it by command-line match is unsafe
-- any shell whose arguments merely mention the script name matches too, which
has now killed the wrong process twice here (once via `pkill -f`, once via a
/proc/*/cmdline substring loop).
`openfut-switch.sh on` prints "the service MUST stay up" -- true, and useless
when nobody is at the terminal. An armed switch pointing at a dead port means
the client hits a closed socket with no fallback.
This turns the documented rollback into an automatic one, failing toward the
Python oracle. The worst case of a spurious trip is a gate needing re-arming;
it can never leave the client broken.
It only ever REMOVES a switch. It does not install one, restart the Rust
service, or touch Python, and it does not re-arm after tripping -- an
unexplained rollback should be a finding to read, not something hidden by
flapping the switch back on.
The probe goes through the switch and speaks TLS, because a bare TCP connect
would succeed against a process wedged mid-handshake.
Tested both directions, not just the happy path: quiet for 45s against a
healthy service, and against a stopped one it failed 3/3 in 9s, rolled back,
and left Python serving -- verified by re-reading all four tables and by which
implementation's log grew.
`verify` inspects `$BIN --identity`, which is the file on disk. That is not
necessarily what is serving. Caught during gate 14 setup: the live process had
been started from 5bc39e9, then `cargo test` re-ran build.rs (the branch ref
moved when an unrelated script was committed) and restamped the on-disk binary
to fc411bb. `verify` then reported "build identity OK" about an artifact that
was not the running service.
`start` now records the stamped commit to $RUNDIR/redirector.commit, and
`verify-running` compares THAT against HEAD, refusing when they differ. The
existing on-disk check stays -- it is the right gate for "may I start this" --
but only the recorded stamp answers "is the thing currently serving the thing I
think it is", which is the question a live gate's evidence depends on.
Two live gates were lost to a second variable I had been asked to eliminate.
The Rust redirector was pointed at the repo's fifa17-recon/tools/redir_cert.pem
(fingerprint F9:16:1A...), while the running container serves a different cert
baked into its image (E7:F9:46...) which the Python redirector, roster and the
rest of the stack all share. So the A/B compared TLS implementation AND
certificate identity at once.
FIFA 17's ProtoSSL caches the server certificate for a backend. The redirector
is the first TLS connection of a session, so its cert becomes the one the client
expects; the next service presenting a different cert fails its handshake. That
is why the redirect itself always succeeded and the failure surfaced later, on
the roster fetch -- "An error occurred downloading the FUT Squad Update".
It stayed invisible because Python's socketserver swallows it: a handshake
failure at accept() raises ssl.SSLError, which subclasses OSError and is
discarded by _handle_request_noblock. No request log, no stderr. Every server
looked healthy while the client could not talk to any of them.
Confirmed on the wire: tls-observe in front of the roster server captured four
ClientHellos from the client, correct SNI and the same 8 static-RSA suites it
offers the redirector, none of which produced a request.
The check is mutation-tested against the real bug: with a redirector started on
the stale repo cert it exits 1 and names the mismatch.
openfut-observe.sh answers the one question no server log can: when a gate
fails and a service logged nothing, did the client try and fail, or never try?
Both look like silence. Two redirector gates were lost to that ambiguity --
"roster server logged nothing" was equally consistent with a broken roster
service, a wrong roster URL, and a client that never asked.
Built on iptables packet counters because this box has no tcpdump, no
conntrack, and no readable kernel log. That last one is verified rather than
assumed: an initial LOG-based version installed correctly and its rules matched
(counters proved it), but the output went nowhere -- journalctl -k has no
entries and dmesg is empty. Counters are also lower volume and record only SYNs,
so no payload can be captured even in principle.
Validated against the live client, not a loopback stand-in: an initial
self-test using this host's own address counted almost nothing, because
locally-generated packets never traverse PREROUTING. Against the real remote
client it counts 8081 at ~4/min, matching the roster server's own log.
Known gap, recorded rather than hidden: the catch-all TOTAL runs well above the
sum of the named ports, so the client makes steady background attempts to ports
not tracked here. It is present during a working session, so it is not the
failure signature, and it is not chased further here.
verify-build-identity.sh now rejects an argument that is not a commit hash.
Passing the binary path instead of its stamp previously produced a plausible
"REFUSING: binary was built from ./target/release/... but HEAD is <sha>", which
reads as a real stale-build finding rather than a caller mistake -- and a
safeguard that cries wolf is one people learn to route around. Usage error is
now exit 2, distinct from a genuine stale build (1) and success (0).
Two live gate attempts failed with "An error occurred downloading the FUT
Squad Update" while the redirect response was verified byte-identical to the
Python oracle. Rolling back to the Python redirector fixed it, so the response
bytes were never the whole contract.
Log archaeology found the discriminator: the client polls
/fifa17/fut/rosterupdate.xml ~4x/min in every successful FUT session, and the
only gap in 300 recorded fetches is 03:47-03:58 -- exactly the two
Rust-redirector sessions. The Blaze RPC sequence over those sessions is
identical (msgNum 0-53), so the divergence is entirely outside Blaze.
A differential lifecycle probe against both redirectors found the two
behaviours this host never reproduced:
* the oracle drains the request body per Content-Length; this host stopped
at the header terminator, leaving unread data in the receive queue, which
makes Linux close with RST rather than FIN
* the oracle holds the connection open ~300ms before closing
(time.sleep(0.3)); this host closed at 0ms
Both are now reproduced. The dwell is a named constant, ORACLE_CLOSE_DWELL,
overridable only so the causal experiment -- set it to 0, confirm the failure
returns -- can be run without a rebuild.
The suite could not have caught either: it sent Content-Length: 0, so there
was never a body to drain. It now POSTs a body, and asserts a split-write body
is fully consumed.
Testing the drain via client-visible symptoms does NOT work -- verified by
mutation: with the drain removed the client still reads the buffered response
and sees close_notify before any reset. So the host records a per-connection
ConnOutcome and the test asserts on that. Both mutations (no-dwell, no-drain)
are now each caught by exactly one test.
This does not yet prove causation for the FUT Squad Update failure; it removes
the only two measured divergences. Gate 6 is the test.
The binary records only the commit it was built from -- no dirty-tree flag.
Cargo will not re-run a build script because another crate's source changed, so
a compiled-in 'clean' claim can be stale and is not a safeguard; that was
verified on the Blaze host.
scripts/verify-build-identity.sh establishes both facts at LAUNCH, where they
cannot go stale: the stamped commit equals HEAD, and the migration crates are
clean. It REFUSES rather than warns, because for a migration gate a warning on
stderr is something to scroll past.
--identity prints the stamp without valid configuration. The launcher must be
able to establish which commit a binary came from BEFORE deciding whether to
run it; requiring a correct environment first would invert the check.
redirector.sh mirrors sidecar.sh: refuses to start with an orphan present or
the port busy, matches the resolved executable rather than the command line
(pgrep -f matches any shell mentioning the name), and stop PROVES the process
is gone and the port free.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
TLS DEPENDENCY, as directed: the openssl crate directly with the `vendored`
feature. NOT native-tls. native-tls abstracts over whatever the platform
provides; here the requirement is the opposite -- precise, evidenced behaviour
for one legacy client -- which needs explicit control of the cipher list,
protocol floor/ceiling and security level. Vendored so a distro libssl update
cannot silently change whether FIFA 17 can connect.
Scoped to this crate alone. Neither OpenFUT Core nor the generic protocol
crates gain an OpenSSL dependency.
CIPHERS driven by the captured retail ClientHello, not by generic legacy
assumptions. The six RSA+AES suites it offers are enabled; RC4 and MD5 are
deliberately NOT, even though the client offers them -- it already negotiates
AES256-GCM-SHA384, so resurrecting RC4 for completeness would weaken the
service for nothing. TLS 1.2 floor and ceiling, matching the observed client;
the floor is not dropped to 1.0 pre-emptively because "the oracle permits it"
is not "the client requires it".
SECURITY LEVEL IS NOT LOWERED. Tried the default policy first, as directed,
and OpenSSL 3.6.3 accepts static-RSA/AES without weakening. No SECLEVEL change
was needed and none is applied; it remains overridable per-listener with
evidence.
CERTIFICATE: the proven Python redirector's material is reused, so the TLS
implementation stays the only variable in an A/B. Verified RSA-2048, CN
winter15.gosredirector.ea.com, cert/key modulus match; the key stays
gitignored.
SHARED CONFIG. New openfut-host-config is now the only crate that reads the
environment, and both hosts resolve endpoints through it. Two hosts each
parsing OPENFUT_ADVERTISE would be exactly the "separate helpers constructing
endpoints from different sources of truth" the address audit forbids.
VERIFICATION BY REAL HANDSHAKE, not by enumeration. The crate exposes no
accessor for a context's configured suites at this version, which turned out
better: the host now rehearses the retail handshake at startup with a client
restricted to exactly FIFA's eight suites and REFUSES TO SERVE if it fails, so
a cipher/version misconfiguration surfaces at boot rather than as an
unexplained failure during a live gate.
Gates 1-5 pass: TLS config unit tests; a FIFA-suite-only client negotiates
TLSv1.2/AES256-GCM-SHA384; each enabled RSA+AES suite negotiable alone; an
RC4-only client is refused; an ECDHE-only client is refused (proving no modern
policy was silently inherited); a full HTTPS round-trip returns bytes
IDENTICAL to the Python oracle's recorded response.
Cargo.lock committed for reproducibility: openssl 0.10.81, openssl-sys 0.9.117,
openssl-src 300.6.1+3.6.3 (OpenSSL 3.6.3). Updating openssl-src is NOT a
routine bump -- it requires re-running the FIFA compatibility gates.
Gates 6-14 need the retail client and are next.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The Blaze switch was hardwired to 42130 and could not intercept the redirector.
Rather than clone it, the iptables logic now lives in one place:
openfut-switch.sh generic: --server-ip --intercept-port --target-port
--name [--client-ip] [--legacy-tag]
blaze-switch.sh thin wrapper, CLI and output UNCHANGED so the validated
gate runbook and sidecar.sh's cross-check keep working
No deployment IP or port literal in the generic tool; 42130 is supplied by the
wrapper, 42127 by the redirector experiment.
VERIFICATION IS INDEPENDENT OF REMOVAL. Rules are created and deleted by their
comment tag; they are verified by parsing the kernel's own FIELDS (chain,
destination, dport, to-ports) with no reference to the comment. Status detects
duplicates, incomplete pairs, conflicting targets under one name, and foreign
redirects on the same port -- which it reports but never deletes. `off` removes
only rules bearing this switch's exact tag, then re-reads the table to confirm.
THREE BUGS FOUND WHILE BUILDING IT, all in the same family as the original
lying rollback:
1. Renaming the tag ORPHANED live rules. Gate 10 deliberately ended with the
switch on, so rules carrying the old tag were still installed and the
renamed tool could not see them -- `off` would have reported success while
traffic stayed redirected. Hence --legacy-tag: a rename must not strand
rules it owns.
2. Deleting by re-feeding the raw `iptables-save` line through the shell fails
on this iptables, which prints `--comment "tag"` WITH quotes; word-splitting
leaves the quotes inside the value so nothing matches. Bare-comment rules
deleted fine, which is exactly what made it look like it worked. Deletes are
now rebuilt from parsed fields and passed as argv elements.
3. `IFS=$'\t' read` collapsed consecutive tabs because tab is IFS *whitespace*,
so an absent `-s` shifted every later field left and produced
`-s <dport> --dport <to_ports> --to-ports ''`. Harmless here, but a shifted
spec that matched a real rule would delete the wrong one. Now uses \x1f.
Mutation-tested against all seven required cases: wrong intercept port, wrong
target port, missing rule, duplicate rule, changed comment representation
(bare vs quoted), and a rollback that leaves a foreign redirect installed --
which exits non-zero rather than claiming success.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The redirector TLS question cannot be answered from the cipher OpenSSL
selected: its server follows client preference by default, so FIFA preferring
static RSA does not prove ECDHE was unavailable. Choosing a TLS stack on that
inference would be a guess. This reads the actual ClientHello.
PASSIVE BY CONSTRUCTION. Bytes relay verbatim both ways, nothing is injected
or rewritten, and the handshake is still terminated by the untouched Python
redirector. A parse failure logs and relays anyway -- observation must never be
able to break the path it observes.
Reports record/client version, supported_versions, SNI, every offered suite by
name, extensions, and a verdict on whether ANY forward-secret suite is offered,
which is exactly the rustls question. Unknown suites print as hex rather than
being dropped.
Verified end to end against the live Python redirector with openssl s_client:
31 offered suites parsed, 18 classified forward-secret, and Python logged the
relayed request and served its 406B serverinstanceinfo -- proving observation
AND pass-through in one run.
Unit-tested on truncated and non-TLS input; the verdict is asserted in both
directions so a static-RSA-only hello reports RULED OUT rather than defaulting
to the permissive answer.
NOTE: that 18-suite result is from openssl s_client, NOT from FIFA. It proves
the instrument works. The actual question is still open until a retail FIFA
ClientHello is captured.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Cheap insurance, explicitly not the real check -- the semantic tests in
deployment_config.rs are what prove propagation, using two TEST-NET addresses
and bind != advertise. This grep only stops the lab subnet reappearing months
from now when the reasoning has been forgotten.
Deployment config legitimately contains real addresses and lives in gitignored
files, so it is never scanned. The frozen baseline doc is allowlisted BY PATH:
it records what a past deployment actually was, and rewriting it would falsify
the record.
Also swapped the lab IP for a TEST-NET placeholder in the usage examples and
error messages of compose/entrypoint/client_arm. Those were already correct
architecture -- every one requires the address via ${VAR:?} -- but using the
real lab IP as the example is the same 'happens to match our lab' smell, and
placeholders keep the tripwire allowlist near-empty.
Mutation-tested: adding a lab address to a source file makes it exit 1.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Mandatory OpenFUT architecture audit. Two real defects found and fixed, plus
the config surface tightened so neither class can recur.
DEFECT 1 -- hidden localhost fallback. The Rust host defaulted POW hosts to
127.0.0.1 while every other URL followed OPENFUT_ADVERTISE, so a remote
deployment would emit loopback POW URLs and fail far from the cause. It also
diverged from the deployed Python entrypoint, which derives them
(POW_HOST="${POW_HOST:-$ADV:8094}"). POW endpoints now derive from the
advertised address; explicit overrides still win.
DEFECT 2 -- Default gave loopback silently. `Endpoints::default()` and
`AdapterConfig::default()` supplied 127.0.0.1, so anything constructing a
config by omission got loopback with no signal. Both `Default` impls are
REMOVED. Loopback is now `Endpoints::loopback()` / `AdapterConfig::loopback()`:
an explicit, greppable decision. Production uses `advertising(host)`.
CONFIGURABILITY. `blaze_port` and `utas_port` are now config, not literals.
The advertised Blaze port is our choice -- the client goes wherever
<serverinstanceinfo> sends it -- and 8099 is the client's own built-in default
but still deployment config. A bad port value is an error, not a silent
fallback to the previous one.
TEST-NET EVERYWHERE. Committed fixtures and tests used the lab's real LAN
address; a test that passes because its constant matches the current lab
proves nothing about relocatability. Redirector fixtures regenerated on
RFC 5737 TEST-NET-1/2/3 plus loopback. Harness scripts no longer default the
client IP to the lab address -- client-state.sh now requires it.
SEVEN REQUIRED TESTS in tests/deployment_config.rs plus host-side coverage:
remote config never silently becomes localhost; missing advertise fails
clearly; bind may differ from advertise; changing the Blaze port changes the
redirect; changing the host updates all 200+ generated URLs with no
stragglers; no helper bypasses central config; mutations are detectable.
MUTATION TESTED, and it found a hole in the audit tests themselves. Hardcoding
utas_base, reverting the POW derivation and re-hardcoding the Blaze port were
all caught. Making the redirector read `bind` instead of `advertise` was NOT:
`advertising()` sets bind == advertise, so the two sources were
indistinguishable. That is the single most likely bypass -- the oracle really
does read bind for nucleusConnect -- so the test now forces bind != advertise
and asserts the bind address never reaches the wire. Re-mutated: caught.
Wire behaviour unchanged: oracle fixtures still current, 153 tests green.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
REDIRECTOR. The first hop's <serverinstanceinfo> XML, byte-for-byte against
the oracle across three advertised addresses. Owns the response only; TLS and
HTTP transport belong to a host, exactly as the Blaze adapter owns dispatch
while the sidecar owns the socket.
The <secure>0</secure> field is the client being told the second hop is
plaintext -- independent corroboration of the plaintext Blaze finding, now
expressed in code.
NUCLEUS IS NOT PORTED, and that is a finding rather than an omission.
Instrumented across every live session:
listener bound YES 0.0.0.0:42131 since 00:21:32
handler logs on connect YES unconditional, before any parsing
client received the URL YES OSDK_NUCLEUS fetched 10+ times
client connected NO zero requests, including 4 full FUT flows
So the long-standing nucleusConnect=0.0.0.0 anomaly is explained: FIFA never
follows that URL on this path. The invalid address has never mattered because
nothing dials it. Porting the stub would add an untested component for no
parity gain.
TLS CONSTRAINT RECORDED, NOT RESOLVED. All 9 observed handshakes negotiated
AES256-GCM-SHA384 = TLS 1.2 with STATIC RSA key exchange. rustls supports only
forward-secret (EC)DHE suites and cannot serve that. Whether the client also
OFFERS ECDHE is unknown -- OpenSSL follows client preference by default, so
preferring static RSA does not prove it is the only option. This must be
instrumented from a real ClientHello before a TLS stack is chosen; the module
docs say so rather than guessing.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Host-side ss cannot see the Python backend's connections: the responders run in
a container, so a client session terminates at 172.20.0.2:42130 inside its
namespace and the host only sees the NAT'd flow. 'ss | grep <client>' on the
host therefore reports nothing while a session is very much alive.
That produced a wrong precondition: 'no .105 Blaze session -- closed' was
reported while FIFA was mid-session on Python, and gate 9 was armed against a
client that had never exited. Python's own log had the answer -- it logs closes
reliably and there was no close for that session.
client-state.sh looks in both namespaces, reports Rust and Python separately,
and exits non-zero while any session is live. An unreachable container counts
as 'cannot confirm', not as 'clear'.
Fourth measurement bug in this tooling, and the most consequential: the other
three mis-COUNTED, this one mis-STATED a precondition and caused an action.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
'pack opens recorded: 45' appeared in the gate 8 report. The UTAS log is
cumulative across the entire deployment, so a bare count reads as if 45 packs
were opened during that gate; the real number was 1.
Now reports both, labelled, windowed from the sidecar's start time (it is
restarted per gate, so that is the gate boundary). A bare count in a gate
report will be read as belonging to that gate, so it has to be the one that
does.
Third counting bug in this tooling: the trace frame counter matched OPEN/CLOSE
markers, the capture and trace were read seconds apart during a live session,
and now this. Evidence tooling gets the same scrutiny as the code under test.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Three fixes, all found while closing out gate 7.
1. Frame count was wrong. It counted lines matching '^conn-', which also
matches the OPEN/CLOSE lifecycle markers, inflating the figure by one or
two. Compared against the capture's record count that looked like a
capture/trace divergence (89 vs 88) when there was none: read at the same
instant, both report 99. Evidence tooling that miscounts is exactly what
this project cannot afford.
2. The raw capture is now copied into the evidence bundle (0600), so a gate's
forensic bytes travel with its report.
3. FUT actions are now observed on the UTAS side. 'Known FUT action succeeded'
is a client-side fact, but FUT actions go over UTAS -- which is never
switched -- so the UTAS log confirms them independently of anyone's
recollection. Gate 7's pack open shows up as:
STORE: opened pack Special Players Pack -> 11 items
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
'Rust did not receive it' is weaker than 'Python did'. The redirector always
runs on Python and is never switched, so it advertises the Blaze endpoint on
every run; whether Python then receives the Blaze CONNECT it just advertised
says where the hop actually went.
This is already visible in the existing logs and settles gate 5-6 more firmly
than the sidecar record alone:
02:02:13 Python REDIR SENT -> 10.10.0.120:42130 (to .105)
02:02:13 Rust conn-0005 CONNECT from 10.10.0.105
Python received NO Blaze CONNECT
Same second, both sides: Python advertised the endpoint and did not get the
connection; Rust did. It is also the mechanism gate 8 needs in reverse.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Committing updates refs/heads/<branch>, not the HEAD file, so watching HEAD
alone left the stamp one commit behind -- observed live, the banner read
a84a72e immediately after 2337431 was committed. build.rs now also watches the
resolved branch ref.
Belt and braces, since cargo still cannot see every source change: sidecar.sh
compares the binary's stamped commit against the tree's real HEAD at launch and
says so loudly on a mismatch. An evidence artefact that names the WRONG commit
is worse than one that names none.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Evidence infrastructure, not protocol functionality. Built before gates 7-10
because those sessions cannot be reproduced -- a later run is a different
session, and the migration-validation runs happen once. Gates 5-6 already went
past without their bytes being recorded.
TWO LAYERS
live FIFA traffic
├── raw capture exact RX/TX bytes, mode 0600, gitignored
└── blaze-sanitize → repository-safe, replayable fixtures
CAPTURE. Off unless OPENFUT_BLAZE_CAPTURE names a file. Deterministic
big-endian container: 20-byte file header, then per-frame records carrying
connection id, a global monotonic sequence, timestamp, direction and the EXACT
frame bytes. RX is recorded as received; TX only AFTER a successful write, so a
record means the bytes were sent rather than intended.
Component/command/msgNum/msgType/payload length are deliberately NOT stored
beside the frame: they are already in its 16-byte header, and a redundant copy
can disagree with the bytes, leaving a reader unable to tell which is true.
Record::header() derives them, so every field the requirements name is
available without duplicating it.
SANITIZER. Redacts only the named tags in SENSITIVE_TAGS (KEY, AUTH, SESS,
MAIL, PML) and reports every substitution with path, kind and length.
Replacement is LENGTH-PRESERVING, so the TDF varint, payload length and Fire2
header are unchanged and the sanitized frame is exactly the size of the
captured one -- asserted per frame, failing rather than emitting a subtly
different conversation. Frames with nothing sensitive keep their exact wire
bytes. Payloads that will not decode are passed through and REPORTED, so a
reader knows they were never inspected rather than assuming they were checked.
TESTS. 39 in this crate. All nine required cases: capture disabled produces no
artefact; RX and TX captured exactly; ordering preserved; fragmented input
(one byte at a time) reconstructs the same frames as a single write; coalesced
input is captured as separate frames, not per-read; capture does not alter wire
output; sanitization removes a real session key from a real captured login;
malformed/truncated/wrong-version captures fail clearly; every listed sensitive
tag is provably reachable.
MUTATION TESTED. Dropping TX capture, truncating captured frames to their
header, and removing KEY from the sensitive list were each verified to turn the
suite red. One mutation was NOT caught: moving the TX capture above the write.
It is indistinguishable while writes succeed and only diverges when one fails.
That invariant is held by code placement and a comment saying so, not by a
test, and the code says as much rather than implying coverage it does not have.
Python oracle unchanged.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The gate 5-6 live run exercised 14 RPCs the recorded fixtures never covered --
Stats, Clubs, OSDKSettings, SponsoredEvents, Messaging::fetchMessages,
Util::getTelemetryServer, Util::userSettingsLoadAll, UserSessions cmd 0x0008,
and the transport PING -- every one taking the empty-reply fallback.
That Python does the same was an inference from reading its dispatch table.
This sends those exact routes to both backends and diffs the replies:
14/14 byte-identical.
Worth keeping: the fixtures were built from what the responder implements, so
they could never have covered what the client asks for and the responder does
not. Only a live session reveals that surface, and this makes it checkable
afterwards.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
For the live FIFA gates. Records switch rules, sidecar status, log, trace and
the Python contract result into a timestamped bundle, and reports CONFIGURED
and OBSERVED state as two distinct sections.
The separation is the whole point. 'blaze-switch.sh status = ON' is an
assertion produced by the same tooling that performs the switch, and that
tooling reported a successful rollback once when none had happened. The
observed half comes from an unrelated source: the sidecar's own record of
which peers connected to it. A non-loopback peer in that log proves the
client's Blaze traffic landed on Rust without depending on reading an iptables
rule correctly.
Verified both ways: loopback-only traffic reports 'a FIFA session did NOT land
here'; a non-loopback peer reports that it observably did.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The compiled-in dirty flag cannot be trusted for this job. Cargo does not
re-run a build script when another crate's source changes, so editing the
adapter and rebuilding the host leaves it reading 'clean' -- verified by
appending a line to the adapter and watching the flag not move.
So the stamp now only names the commit, and the real safeguards run at the
moment they matter and cannot go stale:
* sidecar.sh checks the working tree at LAUNCH and warns.
* check-live-parity.sh REFUSES on a dirty tree, since it produces the
artefact a migration decision is made from. ALLOW_DIRTY=1 overrides for a
throwaway check.
Both scope to the three migration crates, so unrelated submodule dirt does not
trigger them -- a warning that is always on is a warning nobody reads.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
A whole-repo check read DIRTY permanently, because unrelated submodules carry
pre-existing modifications. A warning that is always on is a warning nobody
reads, which defeats the point: the flag exists so a mutated build announces
itself before it can be mistaken for parity evidence.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Prerequisites for the live FIFA A/B. Two safeguards here exist because the
corresponding failure actually happened, not because it was imagined.
BUILD IDENTITY. build.rs stamps commit + working-tree cleanliness; the host
prints commit, tree state, profile and a fingerprint of the bundled config
table at startup, into both the log and the trace. A dirty tree prints an
explicit "do NOT treat results from this binary as parity evidence" warning.
The previous step left four sidecars running, two serving mutated builds, and
nothing in their output said so.
SIDECAR LIFECYCLE (sidecar.sh). start/stop/status/check-orphans/with. Start
refuses when any sidecar is already running or the port is busy. Stop kills,
waits, then PROVES it: PID gone AND port free AND no stray processes, failing
if any check does not hold. `with -- CMD` traps EXIT/INT/TERM so cleanup runs
however the command exits.
Bug found and fixed while testing it: orphan detection used `pgrep -f`,
which matched any process whose command line merely mentioned the name --
including the shell running the test script. It now matches the resolved
executable via /proc/PID/exe. `pgrep -x` is unusable because Linux truncates
the process name to "openfut-blaze-h".
BLAZE SWITCH (blaze-switch.sh). Redirects Blaze to the sidecar with a scoped
NAT rule instead of editing the frozen Python oracle, whose redirector
advertises a hardcoded BLAZE_PORT = 42130. Rules match only <LAN_IP>:42130;
127.0.0.1:42130 is deliberately left alone so Python stays reachable on
loopback and the A/B compares real Python against real Rust. Verified both
directions live: LAN->Rust with the switch on, LAN->Python with it off.
Bug found and fixed: `off` reported success while two rules remained active
and rollback had NOT happened. It matched `--comment "tag"` with quotes this
iptables does not emit -- and the verification used the SAME broken matcher,
so it confirmed its own failure. A rollback that lies is worse than one that
fails. Now matched on the bare tag, verified with iptables-save plus a
tag-independent check that nothing still redirects the port.
Second flaw fixed: `sidecar.sh stop` originally warned about a live switch
and then stopped anyway, creating the exact broken state it warned about. It
now REFUSES, with --force as the deliberate override.
The general rule this all converges on, now stated in the README: a
verification must not share the failure mode of the thing it verifies.
116 tests still passing; clippy clean; Python backend untouched and contract
suite 446/446. NAT table left clean, no orphan processes.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Third migration step, and the one that turns fixture parity into transport
parity. A TCP host that frames a Fire2 stream, keeps one Session per
connection, calls openfut-adapter-fifa17::dispatch(), and writes the returned
frames in order. It owns a socket, a buffer, a session and diagnostics --
that is the complete list. No coins, club, packs, profiles or UTAS logic:
those belong to Core, reached through the adapter later.
NO TLS, and that is evidence-based rather than an omission. The Blaze main
port is plaintext: sending a raw Fire2 Util::ping to the running backend
returns a plaintext PingResponse, blaze_handle uses the raw socket, and only
redir_handle wraps ssl. TLS belongs to the redirector phase.
LIVE A/B AGAINST THE RUNNING PYTHON BACKEND: 101 frames across three
conversations, identical normalized traces. This is the first result in the
migration that is not purely offline. check-live-parity.sh replays the
recorded conversations against both endpoints over real sockets and diffs
volatile-masked traces; session keys and clocks are masked, so anything that
differs is behavioural.
Transport tests cover what fixtures cannot: byte-for-byte replay over a
socket, requests dribbled one byte at a time, several requests in one write,
the four-frame login burst ordered on the wire, session state persisting
across frames and NOT leaking between connections, an absurd payload length
closing the connection instead of allocating, and an undecodable body still
getting a reply. 18 tests here, 116 across the three migration crates.
MUTATION TESTED, including the comparison itself. Dropping a post-login
notification is caught by the probe (frame count) AND the diff; a same-length
content change deep inside a notification body (CTY "US"->"GB", payload 116
both sides) is caught ONLY by the trace digest. So the probe's exit code is
not the test -- the diff is, and the README says so. check-live-parity.sh was
itself verified to exit 1 under mutation.
The listen port is required configuration with no default, so the sidecar
cannot silently collide with the working container. OPENFUT_BIND stays the
advertised-config bind (the adapter derives nucleusConnect from it,
reproducing the oracle) and the listener gets its own setting, so the two are
not conflated.
Gates 1-4 pass and are re-runnable. Gates 5-10 need a FIFA client and are
listed in the README, including the Python -> Rust -> Python -> Rust
back-and-forth that proves the rollback path rather than asserting it.
Python backend untouched and still the live runtime; contract suite 446/446
after this work.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The second migration step: the layer above the codec, deciding WHAT to say
rather than how to encode it. Sits on openfut-protocol-blaze and supplies
what that crate deliberately refuses to know.
blaze/ids.rs component/command/notification tables
blaze/config.rs injectable identity + endpoints, nothing hardcoded
blaze/session.rs per-connection state
blaze/client_config.rs the fetchClientConfig tables
blaze/responses.rs 16 Blaze::* response bodies
blaze/dispatch.rs (component, command) -> Vec<Frame>
Parity is tested, not asserted. fixtures/generate.py drives the real
blaze_responder_v3b.dispatch() and records 49 request->response(s)
transactions, replayed in order against a shared session per connection so
ordering-dependent behaviour is exercised: preAuth captures the locale later
ALOC fields echo, login sets the auth code getAuthToken returns. Comparison
is byte-for-byte including frame count and order.
98 tests green across both crates; clippy clean.
MUTATION TESTED, and it found a real defect in this commit's own design.
Swapping two post-login notifications and flipping one enum inside
AccountInfo both turned the suite red as intended. Hardcoding an address in
utas_base() did NOT -- the config templating substituted raw hosts directly,
making those helpers dead code that merely looked load-bearing. The table now
templates on URL-level tokens ({utas_base}, {nucleus_base},
{pow_content_url}) so they are the single place a URL shape is defined, and
the mutation is caught.
The client config table (227-243 rows per CFID) is generated from the oracle
rather than transcribed: it is reverse-engineered data, not logic, and 400
hand-copied string literals would add a typo class no reviewer can catch. The
generator substitutes real addresses back in and diffs against the oracle for
every section before writing, so the templating is verified rather than
assumed.
Reproduces one known defect deliberately: nucleusConnect is built from BIND,
not advertise, so the live split deployment tells a client on another machine
to reach Nucleus at http://0.0.0.0:42131. Confirmed against the running
container. Reproduced because it is what the only proven-working config does;
fixing it needs live validation and is a separate change. It also implies the
Nucleus stub is not reached in the current remote flow.
Blaze carries no FUT domain state -- no coins, packs, clubs or squads on this
wire -- so Session stays a session key, locale, service name, auth code and a
flag. That boundary will need defending when UTAS is migrated.
Not wired into anything. The crate answers frames; it opens no socket and
owns no runtime. The Python backend remains the live service and the oracle,
and is unmodified (contract suite still green).
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The responder reads OPENFUT_ADVERTISE/OPENFUT_BIND at import time and several
live payloads embed the advertised address (Blaze redirect target, RS4/POW/
roster URLs). Without pinning, regenerating on a machine that exports
OPENFUT_ADVERTISE produces different bytes and --check goes red for a reason
that has nothing to do with the codec.
Pinned to 127.0.0.1 as a placeholder so no real LAN address is baked into a
committed fixture. Not a claim about deployment: remote mode still requires an
explicit advertised address and has no loopback fallback.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
First Rust component of the Python -> Rust migration. Chosen first because
it is the lowest genuinely game-independent layer, it has an executable
oracle, and both existing Rust implementations of it are wrong.
Contents:
* fire2 -- the proven 16-byte frame header, frame/stream splitting
* heat2 -- tag packing, varints, all 11 TDF value types
* message -- frame + decoded body, routed by NUMERIC component/command
* diagnostics -- dumps for capture review
No FIFA 17 command tables, response schemas or notification IDs: this layer
knows 0x0009/0x0007 is component 9, command 7, not that it means
Util::preAuth. That mapping belongs to a game adapter, which is what lets a
future FIFA 18/23 adapter reuse this.
Parity is tested, not asserted. fixtures/generate.py drives the proven
Python responders (heat2.py, blaze_responder_v3b.py) and freezes 56 vectors
-- 31 of them real payloads from the responder's own builders, including
the 11.8 KB preAuth reply. tests/oracle_parity.rs replays every one
byte-for-byte. 54 tests green; clippy clean.
Supersedes two wrong framings, neither of which is removed yet:
* fifa-blaze/crates/blaze-proto/frame.rs -- a 12-byte header with a u16
length, nibble-packed type/options, an error field and a JUMBO flag.
A documented guess at FIFA 23 predating the FIFA 17 recon.
* heat2.py::build_fire2_frame -- packs >IHHHHB3s, msgId at [10:12] and
msgType at [12]. Dead code, but its docstring still states that layout.
Confidence is carried in the types: TypeId::is_verified() reports which
layouts are capture-backed (int/string/blob/struct) and which the oracle
marks UNVERIFIED (list/map/union/varlist/objtype/objid/float), with a test
asserting the unverified ones stay flagged.
Cargo.lock is deliberately NOT included: it re-resolves ~240 lines against
the current registry even without this crate, so that churn is pre-existing
and does not belong in a foundation commit.
The Python backend remains the live runtime and is untouched. Nothing
consumes this crate yet.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
fifa17-recon/tools (authoritative) and fifa17-recon/data now feed the Docker
build directly via the curated runtime-tools.list manifest. The duplicated
fifa17-python/tools+data are removed so the repo has a single source of truth;
the rebuilt openfut-fut-backend:dev image is byte-identical to the previous
deployment (verified: manifest diff empty, 446/446 contract checks pass).
Build context moves from docker/fifa17-python/ up to fifa17-recon/ so the
Dockerfile reads the single-source tools/ and data/ trees. Only the 77 runtime
files listed in runtime-tools.list are installed into /app/tools (baseline image
minus the two git-ignored certs, regenerated in-image). memdump and recon
artifacts are excluded via fifa17-recon/.dockerignore.
The earlier reconcile committed the local working-tree versions of these
files, which are OLDER than the deployed backend. The running container (C)
is byte-identical to docker/fifa17-python/tools (B) and is a strict superset:
it adds profile_path_for/select_account/ensure_security_question (fut_store),
safe_header_for_log/safe_request_path/security_question_route (utas_server),
account_sync_route/_match_call/match_ready_body, plus POW balance fields and
match lifecycle support, with zero unique local functions lost.
Reconciled tree is now a strict superset of B with every shared file
byte-identical; verified via md5 map (0 missing, 0 differing).
- Add 8 files present in docker/fifa17-python/tools but missing from the
top-level tree: fut_accounts.py + 7 test_*.py contracts (all committed in
the server's docker tree; byte-identical to the running image).
- Preserve newer responder work already matching the running container:
utas_server.py (offlineSeason), lsx_responder_v2.py (OPENFUT_BIND),
blaze_responder_v3b.py, autopatch.py, pow_server.py, fut_store.py,
test_fut_contract.py, fifa17-hook-m1.sh.
- Add 30 newer ghidra_queries (draft purchase/state, SBC 9-26, runtime
registries). Local tree is now a strict superset of B with all shared
files byte-identical.
The frozen baseline image predates two hot-patches made in the running
container after build:
* utas_server.py: FUT_MODES-gated offlineSeason block in GetHubData's club
response (keeps the offline-season summary valid)
* test_hub_offline_season_contract.py added to /app/tools
Sync fifa17-python/tools to the running container (verified byte-identical,
237 files incl. the redir cert pair) and snapshot the live FS as
openfut-fut-backend:python-running-2026-08-10 (docker commit). A fresh build
from the committed sources now reproduces the running backend exactly
(baked SHA256SUMS.txt diffed against the container manifest: identical).
Freeze the running offline FUT backend into version control as
fifa17-recon/docker/fifa17-python/ - declarative and rebuildable from a
fresh checkout:
* OPENFUT_BIND / OPENFUT_ADVERTISE client/server split in the responders
(lsx, blaze, roster, utas, pow) + entrypoint.sh; OPENFUT_ADVERTISE is
required for remote mode (compose and entrypoint fail without it)
* docker-compose.yml reproducing the frozen baseline container exactly
(env, ports incl. the 8085->8080 POW-content remap, /state bind, restart)
* .env.example / .env for site config - the LAN IP is never hardcoded in source
* tools/ + data/ staged from openfut-fut-backend:python-baseline-2026-08-10,
verified byte-identical to the running container at freeze time
* client_arm.sh (the 105 client-side arming counterpart)
* Dockerfile bakes /app/SHA256SUMS.txt so any image is self-identifying
* docs/BASELINE-python-2026-08-10.md: frozen image/container/hash record,
restore instructions and rebuild-equivalence procedure
Secrets (redir key/cert, .env) and runtime state (docker/state) stay gitignored.
The live container is untouched pending the .105 launcher audit.
Completed the refusing-modes workflow (ground truth + 4 per-mode investigations +
adversarial verify each + synthesis). All four mode families -- Seasons, Draft,
SBC/Objectives, Tournaments -- are NOT_SERVER_REACHABLE, HIGH confidence, all four
adversarial refutations failed.
Live re-confirmed on pid 24653 (slide proven via FNV control): every named
mode-gating byte reads ENABLED=1 (IS_FRIENDLY_SEASON_ENABLED +0x1fd3a,
IS_TOURNAMENT_QUIT_ENABLED +0x1fd3b, IS_DRAFT_MODE_ENABLED +0x1fd3d, plus the
unnamed offline-draft-enable +0x1fd3e) yet the tiles stay greyed.
The new lead this pass added -- do the six /hub mode sub-objects gate availability?
-- is refuted: friendlySeason/offlineSeason/onlineSeason/draftSummary/tournament/
tournamentProgress carry only stats and display strings, no enabled/available/
unlocked atom. They are cosmetic, exactly like hub.tradePile. The one
server-writable input that exists (friendlySeasonsEnabled -> +0x1fd3a via applier
FUN_18011dc50) has its sole reader in the packed FIFA17.exe front-end via a vtable
getter with no CardsDLL caller, and it is already 1. The refusal is decided in the
Denuvo-packed Frostbite front-end, which has no server surface.
docs/plan-2026-08-06-refusing-modes.md: full evidence chains, gate-byte table, the
six sub-deser field maps, per-mode verdicts.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Lrx9to3pihN6Sm9sXgc8np
The Transfer List hub tile read "0 items / Selling 0" while a card was actively
listed. Enumerating the /hub parser FUN_180139610 straight from the on-disk
CardsDLL (objdump) refutes the old ENDPOINT_MAP claim that it uses C++ reflection
with "no atom ladder, nothing to enumerate": it has an ordinary running-sum atom
ladder reading 18 atoms. The tile is fed by hub.tradePile (0x333), a nested object
(sub-deser 0x18013ead0) reading count/selling/sold as scalar ints -- the same
scheme as GetAuctionCount, so serving it in the hub body is freeze-safe. The tile
never re-polls the standalone /tradePile/counts, which is why fixing that endpoint
alone did not move the tile.
Also: the hub tile polls LOWERCASE tradepile/counts while the Transfer List screen
uses camelCase tradePile; our case-sensitive routes matched only the screen, so the
tile's counts call fell through to /trade and got a shape the counts deser skips.
Made the tradePile routes case-insensitive.
And bake the proven transfer-market flags (FUT_TRADING/PILESIZES/TRADEABLE/
DISCARD_TABLE/DISCARD_SEND) into openfut-fut.sh so a plain `start` brings up the
working state instead of regressing trading to greyed-out.
- tools/utas_server.py: hub_data() serves tradePile:{count,selling,sold};
tradePile routes now re.I
- tools/openfut-fut.sh: utas launched with the working flag set
- docs/ENDPOINT_MAP.md: full 18-atom hub map + tile map, correction of the
reflection claim
- tools/ghidra_queries/objdump_atom_ladder.py: the objdump-based atom-ladder
decoder used to derive the above
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Lrx9to3pihN6Sm9sXgc8np
Two follow-ups on the working transfer market.
1. GET /tradePile/counts now returns the FutGetAuctionCount shape
({count, maxAuctionsAllowed, offered, selling, sold}, all scalar ints, atoms
0xbc/0x1bf/0x1e5/0x2b8/0x2c9) via a dedicated route ordered before /tradePile.
Previously it fell through to tradepile_route and got the auction-LIST body, which
the counts deser skips, leaving every tally at its constructor default. Survivable
but wrong; the doc flags the loaded byte at +0x28 as gating a completion-handler
branch. selling reflects real STORE.listings().
2. Narrowed the marketdata bare-array fix to /pricelimits only. The client sends TWO
marketdata requests: /marketdata/pricelimits (GetSuggestedPricing, a bare array,
the thing that froze) and plain /marketdata?defId=N (price comparison, an OBJECT).
The prior commit returned the array for both, which the contract suite caught
(test_market_bodies: 'list' has no attribute get) -- plain /marketdata wants
{minPrice,maxPrice} and was never the freeze. Returning the array for it would be
the same desync in reverse. Now: pricelimits -> array, plain marketdata -> object.
The contract suite catching my over-broadened fix before it reached the game is the
suite doing its job. 439 contract checks pass, market unit suite passes.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The subsystem that was fully greyed-out this morning now lists a card on the transfer
market: price screen, Submit, "your item is now up for trade", TRANSFER LIST 0/100,
auctionCount 1, and STORE.listings() holds the auction. Every step verified at the
instruction level first, then confirmed live. Three fixes, all behind flags, all off by
default until this run proved them.
1. WE WERE BANNING OUR OWN TRADING. userInfo.feature (atom 0x11c) is a RESTRICTION map,
not a grant; we sent feature={"trade":true}, which is a trade BAN. Verified in
q_feature_trade.py: FUN_18013ec10 parses feature/trade into userInfo+0x17c, and at
the massinfo END_OBJECT the client runs
cmp byte [rsi+0x17c],0 / jz skip / mov dword [rsi+0x50],0
feeding applier 0x18011dc91 -> IS_TRADING_ENABLED (model+0x1fd2e) = 0. It runs LAST
and unconditionally, which is why the gate read 0 all day regardless of /settings or
the Blaze config store. FUT_TRADING sends feature={} instead. Live: gate flipped
0 -> 1 on UT re-entry (model rebuilt, pointer changed, byte read 1).
2. TRANSFER LIST CAPACITY 0/0. pileSizeClientData (massinfo atom 0x227, parser
0x18013adb0) is the capacity, NOT the "MY CLUB counter" the old comment claimed.
Verified in q_pilesize_keys.py: exactly two storing arms, key 2 -> model+0x1fd1c
(TRADE_PILE_SIZE) and key 4 -> +0x1fd20 (watch list), every other key SKIP'd. The old
code would have sprayed the 246 club count into the capacity. FUT_PILESIZES sends
key 2 = 100, key 4 = 50. Live: capacity read 0 -> 100, header showed 0/100.
3. THE PRICE SCREEN FROZE THE CLIENT. GET marketdata/pricelimits was answered with an
OBJECT {minPrice,maxPrice}; the deser 0x180163ee0 reads a BARE TOP-LEVEL ARRAY
(root loop while tok != 0xd), so object-where-array desynced the SAX reader into the
0x1801c7f1a busy loop (confirmed live: utime climbing 227 ticks/s, core pinned).
Verified in q_pricelimits.py: element fields defId 0xcf, maxPrice 0x1c2, minPrice
0x1ca, all scalar ints. marketdata_route now returns a bare array, one element per
requested defId. Live: price screen opened and Submit succeeded.
Corrected along the way, all now in the code: two prior "trading root causes" from
earlier today were wrong (the Blaze IS_TRADING_ENABLED keys are output-only names, and
the applier is a virtual method at vtable+0x988, not unreachable). Those refutations are
recorded in blaze_responder_v3b.py and the doc.
Also lands the transfer-market recon doc (plan-2026-08-06-transfer-market.md) and the
market Ghidra query set.
Server-authoritative economy note: the 5% transfer fee and the price bands (currently a
150..15000 placeholder per defId) are not yet real; that is refinement, not a freeze.
The live-auction market SCREEN ("List on Transfer Market" browse) is a separate surface
still to do (P4 auction-counts route, P5 empty market bodies).
Live: 439 contract checks pass. Card listed and persisted, auctionCount 1.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Ships the club-item subtype correction and the tradeable plumbing, and records two
refutations of claims made earlier in the same session. Nothing here is a working fix
for trading; the honest state is that the gate is still closed and we now know more
about why.
REFUTED 1: "the Blaze client-config store opens the trading gate". It does not, and the
flag is inert. IS_TRADING_ENABLED is an OUTPUT NAME. FUN_18006cc60 is a publisher: at
0x18006ccc6 it calls [rax+0x270] to READ gate byte 0x1fd2e, then lea rdx,[
IS_TRADING_ENABLED] and hands the value out under that name. The only rip-relative
reference to the literal 0x1801fc118 in all of .text is that lea; there is no comparison
against it anywhere, so no client-config key of that name can be read as an input. That
also undermines the IS_* store keys shipped beside it: their apparent success was never
actually attributed to them.
REFUTED 2: "the gate byte flipped to 1". It reads 0. It was measured as 1 shortly after
CardsDLL mapped and that was over-claimed as a success; a thorough re-measurement read 0
on the SAME pid and model pointer, and a fresh session reads 0 with an unambiguous raw
dump (model+0x1fd18.. = 01000000 00000000 00000000 00000000 3c000000 01 01 00 01, the 00
being 0x1fd2e). Either the first read was transient or something clears it after login.
The only writer is FUN_18011dc50 at 0x18011dc91, so a 0 means something RAN and wrote it.
AND THE "/settings IS DEAD" CLAIM FALLS TOO. FUN_18011dc50 is not unreachable: it is a
VIRTUAL method at model vtable slot +0x988 (absolute pointer 0x18021cc28). A direct-call
search found no callers because Ghidra does not resolve virtual calls, which is the same
dispatch-form trap that has now produced seven wrong verdicts here. The real chain is
settings response -> FUN_180174630 -> FUN_18013c6d0 (deser)
-> completion callback FUN_180173e00 -> vt+0x988 and vt+0x998 -> gate bytes
and FUN_180173e00 bails before applying anything unless the int at response+0x1c is
zero. Which atom writes +0x1c is unknown and is the thing worth chasing.
The measurement behind that claim also had a gap: it checked +0x1fd14, +0x1fd4c and
+0x1fd54 for the maximumTradePileSize=77 probe but NOT +0x1fd1c, which is the actual
TRADE_PILE_SIZE (read via vt+0xa58 = FUN_18011bf30). So the probe never tested the field
it needed to. Serving 77 and reading +0x1fd1c is the clean falsifier and is still open.
Recovered and worth keeping: an authoritative slot-to-name table from the publisher.
vt+0x270 IS_TRADING_ENABLED -> +0x1fd2e vt+0x2b0 IS_FRIENDLY_SEASON_ENABLED -> +0x1fd3a
vt+0x2b8 IS_TOURNAMENT_QUIT_ENABLED -> +0x1fd3b vt+0x2c0 IS_PROCESSING_STATE_ENABLED -> +0x1fd3c
vt+0x2c8 IS_DRAFT_MODE_ENABLED -> +0x1fd3d vt+0x2d8 IS_STORY_MODE_REWARD_ENABLED -> +0x1fd3f
vt+0x2f0 IS_RETURNING_USER_REWARDS_SCREEN -> +0x1fd40 vt+0xa58 TRADE_PILE_SIZE -> +0x1fd1c
That also locates the red TRANSFER LIST 0/0: it is +0x1fd1c, currently 0.
WHAT IS ACTUALLY SHIPPED HERE, all default off:
* FUT_TRADEABLE sends untradeable=false. Verified landing at item+0x49 (stored
INVERTED by case 0x361) on a live club record. Applied on every READ path, not only
in _item(), because the save holds 246 items minted before the flag existed and the
club route serves them straight from the save. That gap was caught by reading the
served JSON, not by unit-testing the factory.
* FUT_TRADING adds tradingEnabled and IS_TRADING_ENABLED to the Blaze config. Kept
only as a record of the refutation, with the reasoning inline so nobody retries it.
* fut_clubitems FAMILIES subtypes corrected: kit 9, stadium 10, badge 11 (cardtype 7,
not 9), ball 30, league logo 31. Every previous value sat in the 0x91..0x96 TROPHY
block. probe_shelf's candidate set lacked 9, 10 and 11, so the probe route the docs
preferred could never have answered this for three of five families.
* Club kits and badges now carry teamid, reintroduced ALONE after the 2026-08-05 crash
(which was never bisected; value is the established suspect and that response also
carried 30 items across five wrong subtypes). itemType dropped: it was unobserved and
never copied into the record.
Live: 439 contract checks, 414 card-family checks, market suite, all pass. The transfer
market still refuses with zero requests and the menu entries are still greyed.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Card-subsystem pass, 11 agents plus three adversarial verifiers. Full writeup in
docs/plan-2026-08-06-card-subsystem.md. Two of the results below correct things I
committed earlier today.
THE GREYED-OUT TRANSFER OPTIONS ARE EXPLAINED. "Place on Transfer List" and "List on
Transfer Market" have been disabled in the reveal screen and nobody knew why.
TO_TRADE_PILE (FUN_1801a7260) requires BOTH item+0x49 tradeable AND a service gate at
vtable slot +0x270. That slot is `movzx eax, byte [rcx+0x1fd2e]; ret`, and 0x1fd2e is
the tradingEnabled gate byte. Read live and reproduced independently:
slot +0x2b0 friendlySeasons disp 0x1fd3a VALUE=1
slot +0x2c8 draftMode disp 0x1fd3d VALUE=1
slot +0x2e0 packOpeningAnim disp 0x1fd45 VALUE=1
slot +0x270 tradingEnabled disp 0x1fd2e VALUE=0
tradingEnabled is the FIRST gate byte found that is not 1. This partly rehabilitates
the settings work from this morning: that plan died because every gate it targeted
already read 1, and the conclusion drawn was that the settings array does not matter.
It does. It matters for a flag nobody was looking at, and tradingEnabled is ALREADY in
_SETTINGS_KEEP, plumbed and never sent because _SETTINGS_MODE defaults to off.
So the fix is two things, not one: FUT_SETTINGS=keep AND untradeable false. Shipping
only the boolean would look like the finding failed.
THE DISCARD "MISS" NEVER EXISTED, which corrects e3092ca. fcc_discardcoins is resident
and complete, the client lookup runs and is correct, and it lands at item+0x3c. The
tile simply binds +0x38, which is OUR value, and nothing falls back to +0x3c. So the
client was not failing a lookup; it was faithfully displaying the 0 we sent. Same
observable, completely different mechanism, and the version in e3092ca is wrong.
FUT_DISCARD_SEND remains exactly the right fix, now for the right reason.
WHAT FUT PAYS FOR STAFF IS NO LONGER UNKNOWN. Same formula, but the rating input is the
table `value` column: gkcoachcards 9000081 value 66 gives 36, and the client's own
+0x3c reads 36. That closes the gap I flagged in e3092ca as not-guessed.
CLUB ITEM SUBTYPES, the standing unknown in CARD_SYSTEM.md, are settled: kit 9,
stadium 10, badge 11 are cardtype 7 (not 9), ball 30, league logo 31 by elimination.
All five constants in fut_clubitems.FAMILIES are wrong and all five currently sit in
the TROPHY block 0x91..0x96. Note the probe route the doc preferred could never have
answered this: probe_shelf()'s candidate set lacks 9, 10 and 11, so it would have spent
a launch and returned nothing for three of five families.
THE CARD MODEL FIELD MAP now exists, 28 rows, every field we send with the byte it
lands on and whether the client keeps it. Built by diffing what we serve against the
parsed records in the live heap (stride 0x180, anchored by a satellite back-pointer
rather than by assuming the +0x38 offset). Corrections that change what we serve:
+0x54 is the discard LEVEL not itemType, +0x49 is untradeable INVERTED, +0x5c is
itemState, definitionId is not an atom at all.
A HIGH-CONFIDENCE ABSENCE CLAIM WAS REFUTED IN VERIFICATION: playStyle IS stored, at
+0x88. Its controls were raw scalars while playStyle is a DECODED scalar, so the
control was the wrong FORM. That is a new variant of the absence trap, which has now
cost six wrong verdicts, and it is recorded in the doc.
FIX TO MY OWN PATCH from e3092ca: purchased() and last_pack() lacked the _with_discard
wrapper that items() had, so the pending pile, which is the one place a quick-sell
value is actually read, served unstamped cards. Found by verification, not testing.
All three read paths now stamp.
Correcting an overstatement in e3092ca: "turning the flag off is a true revert" holds
for the read paths, which copy, but NOT for cards minted while armed, because _item()
stamps at creation and those persist (9 items currently). Kept deliberately: the pack
reveal serves itemList straight from open_pack(), not through purchased(), so removing
creation-stamping would leave the screen that matters unstamped. Persisted values are
correct and self-heal, since every read recomputes and overwrites.
Nothing here has been on screen. Six patches are proposed in the doc as pasteable text,
env-flagged, defaulting off, none applied.
Live: 439 contract checks, 414 card-family checks, market suite, all pass.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Follow-on to 21a81ad, both halves confirmed live.
FUT_DISCARD_TABLE IS PROVEN. Quick selling a 75-rated rare gold paid 600 and the
balance moved 9,844,900 -> 9,845,500, exact. The invented tier would have paid 150.
75 * 800 / 100 = 600, straight off the recovered fcc_discardcoins row.
BUT THE SCREEN SAID 0, which is why this commit exists. The value was right and
invisible: "Quick Sell 0" on the card and "Quick Sell all remaining Items 0" too, so
the wallet contradicted the display on every card. Cause is the guard the table work
had already reversed. FUN_18013fe00 stores our discardValue (atom 0xd7) at item +0x38
and 0x180141025 skips the client's own fcc_discardcoins lookup only when that value is
NON-ZERO. We seeded 0, so the client ran its own lookup, that lookup returns no row for
our cards, and it rendered 0.
FUT_DISCARD_SEND puts the value on the wire and the client uses ours verbatim. Live
result, one launch:
a 77-rated rare gold shows "Quick Sell 616" (77 * 800 / 100 = 616, exact)
"Quick Sell all remaining Items" shows 5,640
and 5,640 is exactly the sum of the ten rated PLAYER cards in the pending pile. The
eleventh, a consumable, is excluded by the client from the bulk figure; the twelfth is
a staff card we deliberately send no value for. Both halves of the display now agree
with what the server credits.
WHY THE CLIENT'S OWN LOOKUP MISSES for our cards is still UNKNOWN. This routes around
that question rather than answering it, and it is worth answering.
TWO CORRECTNESS FIXES THE REPORT DID NOT COVER, both found by running the whole save
through the formula rather than trusting the 22/22 sample:
* The recovered formula scales by rating, so a rating-less STAFF card collapses to 0.
The old tier paid 50, so shipping it as-is was a regression. discard_value() now
returns None when the formula does not apply and callers fall back. What FUT really
pays for staff and consumables is UNKNOWN; the likely answer is the unscaled table
price, but that is a guess and is not shipped as one.
* A missing table row also returns None rather than 0, for the same reason.
Verified across all 246 club items plus the pending pile: not one pays 0 coins.
discardValue is stamped inside the single item factory so every path gets it (pack
contents, starter grant, club, market), and additionally on the club READ path, because
_item() only covers cards minted from now on while the save already holds 246 built
before the flag existed. Stamped on the way out and NOT persisted, so the save stays
clean and turning the flag off is a true revert. Confirmed: 0 items on disk carry the
key.
FUT_DISCARD_SEND requires FUT_DISCARD_TABLE and silently stays off without it, so the
invented tier can never reach the screen and become authoritative-looking.
Freeze risk: low and in the safe direction. discardValue is a plain INT read by the
scalar getter 0x1801c79d0; every freeze on this project has come from an object or
array where a scalar was expected, never the reverse.
Both flags still default OFF. Live: 439 contract checks pass, market suite passes.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Multi-agent pass over the store subsystem, 11 agents, findings run through three
adversarial verifiers. Full writeup in docs/plan-2026-08-05-store-subsystem.md.
THE REAL DISCARD TABLE IS RECOVERED. quick_sell() paid an invented rating tier
(600/300/150/50) that was wrong for every single card. The real table is
fcc_discardcoins in the client's own game DB, 141 rows keyed (cardtype, level, rare),
read out of the running client and verified 22/22 against live items:
value = round_half_up(rating * price / 100)
level = 3 if rating >= 75, 2 if 65..74, else 1 (0x180141e8a..0x180141ea3,
derived from rating, NOT a wire field)
cardtype = FUN_1800d8330(cardsubtypeid), decoded from its jump table and checked
across every subtype 0..599 with zero disagreements
A 94-rated gold rare is 752, not 600. A 76 rare is 608, not 150. A 55 bronze is 17,
not 50.
This also closes a disagreement nobody had noticed: the CLIENT already computes and
displays the correct value locally whenever our discardValue (atom 0xd7) is 0 or
absent. FUN_18013fe00 stores our value at item +0x38 and the guard at 0x180141025
skips the local computation when it is non-zero. So the screen has been showing the
real number while the server paid a made-up one, on every quick sell ever made.
Verified beyond what the report claimed, because a missing table row pays ZERO and
that would be a regression the old flat tier could not produce: across all 236 items
in the live profile, 230 map to cardtype 1 and 6 to cardtype 6, and NOT ONE would pay
0 coins. Table reproduces at 141 rows and the worked example lands exactly.
ZERO WIRE CHANGE, FUT_DISCARD_TABLE default off. Nothing new is sent; only the coin
figure the server credits moves. This is the patch worth defaulting on after one
in-game check, which is simply quick-selling a card and seeing the coins paid match
the value the card was already displaying.
THE GROUPING BUG IS NOT IN CARDSDLL, and the fix ranked first would have wasted a
launch. Live in the running client all three display groups own exactly the right
pack, there is exactly one copy of each pack record in 4 GiB, and nothing we send is
mis-parsed. The parsed model is correct and the Scaleform layer picks the wrong pack
when turning a tile click into a category id. displayGroupAssetId is served as 1/5/6
while the screen's category field reads 3, and group tiles carry a hardcoded
CATEGORY_ID of 0. Confirmed by direct read: ordinal 3, assetId 6, i.e. Premium, while
the last click was Gold.
The heap map that made this possible, all scoped to one pid: display-group vector
control block, 3 elements of 0x108; group record fields at +0x00 sortPriority,
+0x04 displayGroupAssetId, +0x40 a one-element pack vector; inner pack record 0x1a8
with packType at +0x38, ids at +0x70/+0xac, price at +0xa0, quantities at +0xc0..+0xd0.
extPrice SHOULD BE DELETED, not corrected. Both sub-parsers read only
externalPriceId; amount and currency are discarded. Sending the key at all creates an
"mtx" currency row that switches on a real-money price line the client can never fill
offline, which is the literal "or %1s" on every tile.
A WORRY NOBODY HAD RAISED, and I confirmed it from our own logs: the client has sent
packId 6 on every purchase it has ever made, four for four tonight and six for six
across history. We have never observed a successful buy of anything but Premium Gold.
Also settled: FUT_STORE_DISPLAYGROUP=0 is the right resting state, argued from
mechanism rather than from history; FUT_USERINFO=packs stays off because the
unopened-pack counter is client-mutable and the flag ladder silently drops squadList;
POST /user is a latent hard freeze that has never fired because the client never
issues that POST.
Honest coverage: the ActionScript layer is unread by everyone and every remaining
store mystery lives there.
Live: 439 contract checks pass, market suite passes, both flags off.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Live run, 2026-08-05 evening. Three results, one of them a route that has been dead for
the whole life of the project.
QUICK SELL WAS NEVER SERVED. Captured on the wire:
DELETE /ut/game/fifa17/item/100000240 (single card, id in the URL, no body)
ENDPOINT_MAP documented the path as `ut/delete/game/%s/item`, ROUTES was built from the
doc, and the regex therefore never matched a real quick sell. Every quick sell fell
through to the catch-all, so quick_sell_route() and STORE.quick_sell() behind it had
never once been called.
The empty response was not a harmless no-op. This body is BALANCE-BEARING: the client
takes its coin total from it, so with totalCredits absent it rendered an uninitialised
value. A real session showed 1,133,686,384 coins against a true balance of 9,889,600.
Display artifact only, corrected by the next GET /user/credits, and the save was never
touched, but it is the reason a stub is not acceptable here.
quick_sell_url_route() serves the real form and returns the corrected shape,
{"items":[{"id":N}],"totalCredits":N}. DEFAULT ON, which the house rule now permits:
two quick sells fired through it in one session, each credited 150 and removed the
card, and the coin arithmetic reconciled exactly against the 15,000 pack purchases
either side of them. An unknown id returns {"items": []} rather than claiming a sale we
cannot account for.
Still UNKNOWN and deliberately not chased: whether the client ASSIGNS totalCredits as
the new balance or ADDS it as a delta. We send the new balance. The discriminating
window is about five seconds wide, because the client refetches GET /user/credits
straight afterwards, so either reading self-corrects and the practical impact is a brief
wrong number. It mattered only while we answered {}, because that garbage persisted.
The credit amount is STORE.quick_sell()'s invented rating tier, not FUT's real discard
table, which remains unknown.
finalFunds IS THE RENDERED COIN PRICE. Served funds=15000 / finalFunds=4321 on one pack
and the tile read 4,321. funds is not displayed. ENDPOINT_MAP updated to CONFIRMED LIVE
with the method recorded. FUT_PRICE_PROBE, the flag that produced it, stays default off
and is disarmed: it puts a price on a tile that the buy path does not charge.
TWO UNPLANNED FINDINGS, both recorded for the next round rather than fixed here:
* The store grouping is broken and it is NOT cosmetic. All three packs collapse into
one display group, and the Bronze, Gold and Premium group tiles all drill into the
same single Premium Gold pack, so TWO OF THREE PACKS CANNOT BE BOUGHT. We send
displayGroup {"value": name} but never displayGroupAssetId (0xda), so everything
lands in group 0. The docs had this parked as a cosmetic "tiles read unknown"
issue; it is an availability bug.
* The FIFA Points price renders as the literal "or %1s", an unsubstituted printf
placeholder. extPrice.finalPrice is served as {"amount":N,"currency":"mtx"} and
"mtx" is evidently not a currency token the client resolves. Cosmetic.
Corrected in passing: packContentInfo DOES reach the tile (11 ITEMS / 11 GOLD /
11 RARES against exactly what we serve). An earlier screen showing zeros was the
display-GROUP level, which carries no content info. D3 was right.
Live: 439 contract checks pass, both probe flags off, store prices back to honest.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Three things: the envelope rule was wrong and is corrected, /hub is settled, and two
documented response shapes that would freeze the client are fixed.
THE CORRECTION. The previous commit concluded that a three-token root consumes `{`, the
first field name and that field's value without dispatching them, so the first key of a
flat body was silently eaten, and that `login` had therefore never been delivered on
POST /user. That is WRONG and is withdrawn, along with the claim that the key order of
the auth dict is load-bearing.
The first call to FUN_1801c7f10 returns token 7 and consumes NO input. It is a
once-only start-of-document token, guarded by the flag at parser+0xda together with the
zero character counter at parser+0x30. So the three tokens are BOF, `{`, and the FIRST
FIELD NAME, and the key loop dispatches from that first key onward. The `== 10` test on
the third token is not an envelope check, it is the empty-object early-out: for `{}` the
third token is END_OBJECT and the root exits with its constructor defaults intact, which
is why answering `{}` has always been safe.
Corrected enum: 7=BOF 9=START_OBJECT 10=END_OBJECT 11=FIELD_NAME 12=START_ARRAY
13=END_ARRAY. The enum itself was right before; the inference from it was not.
HOW IT WAS CAUGHT, which is the part worth keeping. Not by more decompiling. /hub is
served flat and the wrong model predicted its first key would be discarded, so the
prediction was checked against the client's own memory: clubPlayers read back as 205,
the value the server sent, at model+0x1fd70+0x3c with the slide proven against the FNV
prologue first. One live read refuted a chain of otherwise sound static reasoning in
about a minute. tools/hub_counter_probe.py keeps it repeatable.
Consequence worth flagging: a wrapper is not just unnecessary for these roots, it would
be harmful, since a wrapper key hashes to an atom with no arm and the whole object is
skipped. That makes the createPackResponse envelope DOUBTFUL rather than confirmed.
Atom 0xbe has no arm in FUN_180162880. There is no live evidence either way because
nothing has ever parsed that body, so the buy path is left exactly as it is.
TWO ERRORS OF MINE ON THE WAY, both recorded in the doc because both are cheap to
repeat. I searched for RS4:FutGetHubServerResponse, found nothing and reported that no
hub class existed; the class is FutGetHubDataServerResponse (literal 0x18022ce40,
vtable 0x18022cd48, deser 0x1801738b0, control FutSquadSave -> 0x180171a60 matched in
the same run). Then I scanned 152 deserializers for clubPlayers, got zero hits and a
passing control, because the guard is `!= 0x90` and my pattern only matched `== 0x`.
The control passed only because auctionCount happens to use `==`. A control that does
not exercise the same code shape as the target is not a control. The comment already at
utas_server.py:1076 had the hub chain right the whole time.
ENDPOINT_MAP corrections, both freeze-risky as written, neither affecting what we serve
today:
* duplicateItemIdList is an ARRAY OF OBJECTS (element deser 0x180138e10), not the int
list at :1095. Bare ints where the element parser expects objects is a tokenizer
desync, i.e. a hard freeze at 0x1801c7f1a. Control that this is not a misread:
dreamSquads 0xe9 in FutMoveCard genuinely is a bare int array.
* FutDiscardCardServerResponse is {"items":[{"id":N}],"totalCredits":N}. There is no
top-level id.
No behaviour change. utas_server.py is comment-only. 439 contract checks pass.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The open question was whether a response deserializer DESCENDS a wrapper or PROBES for
one. Three roots spend an identical three tokenizer calls before dispatch, yet we serve
some bodies wrapped and some flat, and not all of those could be right.
TOKEN ENUM, decoded from the class table at DAT_18023dd40 and the switch in the
classifier FUN_1801c67a0 (the push/pop arms key off container state 2 = object,
3 = array):
9 START_OBJECT case 0x64, pushes state 2
10 END_OBJECT case 0x65, pops state 2
11 FIELD_NAME confirmed independently: FUN_18013bd40 tests +0xd0 == 0xb then
atom-hashes the string at +0xf8
12 START_ARRAY case 0x66, pushes state 3
13 END_ARRAY case 0x67, pops state 3
1 error the caseD_78 sink
So the three tokens are `{`, the first FIELD_NAME, and the token opening that field's
value. The envelope is structurally required and its name is NEVER hashed, which is why
FutCreatePack's ladder has no arm for createPackResponse (0xbe) and does not need one.
Coverage for that absence: the ladder has exactly four arms (0xec, 0x16e, 0x1dd, 0x264)
and 0xbe does not occur anywhere in the full 4702-char decompile, printed in full.
The competing reading rested on a factual error. It claimed the /purchased root spends
the same three tokens. FUN_180124ee0 spends TWO and hands off to FUN_18013bd40, which
spends the third. Same total, split across two functions. /purchased never was a
counterexample.
THE BUG THIS FOUND IS NOT THE ONE THAT WAS PREDICTED. The doc expected starterPack,
squad and userData to be swallowed on POST /user. They are not. FutCreateUser
(0x18014cc60) has ladder arms for exactly the five keys we send, and four of them
dispatch correctly at the outer level. The one that does not is `login`: its name is
eaten as the anonymous envelope and its value as the third token. It has an arm, so the
client wants it, and it has never once been delivered.
The second-order consequence matters more than the first. The key order of that dict is
load-bearing and nothing said so. Put userData first and the client loses the entire
user record, silently, with no error and no log line. That warning now sits in the code
next to the dict, which is the only place someone about to reorder it would look.
No behaviour change here. The utas_server.py edit is a comment. 439 contract checks
still pass. The probable proper fix, wrapping all five keys one level down inside a
single envelope key, is a hypothesis with a mechanism rather than a proven fix, and it
touches the login path, so it is not made here and would go behind a flag defaulting
off.
Writeup is section 2 of docs/plan-2026-08-05-pack-opening.md, added by the previous
commit. Opened by this and still UNKNOWN: GET /hub is answered with a flat two-key
body, which under this rule a three-token root would silently truncate, but there is no
FutGetHubServerResponse class and neither atom has a code xref, so /hub may not go
through a generated root at all.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
A twelve-agent pass over the parts of pack opening we did not understand, run against
the live client (CardsDLL slide proven, not assumed) plus static CardsDLL. Findings
below survived an adversarial verification round that corrected several of them; where
a verifier and a finder disagreed, the verifier won.
THE HEADLINE IS A NEGATIVE, and it deletes work rather than creating it. There is no
pack-inventory endpoint in FIFA 17 and there never was. Proven three independent ways:
the 48-entry UTAS route template array at 0x18021df80, a regex for "ut/" over the whole
PE, and the 125-row client action table at 0x1802caa20, which is the complete set of
requests the client can originate. "Serve the pack inventory" comes off the backlog.
The unclaimed-pack tile and My Packs are two fields on responses we already build.
Corrections to ENDPOINT_MAP.md, both freeze-risky as written:
* duplicateItemIdList is an array of OBJECTS (element parser 0x180138e10: itemId
0x16d, duplicateItemId 0xeb, itemLoans 0x16f, duplicateItemLoans 0xed), not the
int list documented at :1095 and :218. Control that this is not a misread:
dreamSquads 0xe9 in FutMoveCard genuinely is a bare int array and parses with no
inner object loop. We serve [], so this is a docs bug today and a live freeze the
moment somebody implements it from the map as written.
* FutDiscardCardServerResponse is {"items":[{"id":N}],"totalCredits":N}. There is no
top-level id. :968-971 is wrong twice over.
packContentInfo is DECORATIVE. It is read only into a store-tile view model, and
nothing compares the declared counts against the delivered itemList, so open_pack()
does not have to honour the distribution.
The reveal is entirely CLIENT-SIDE. Walkout, tiering, colours and ordering are
arithmetic over fields we already send. Genuine outstanding server work reduces to
three items: duplicates, quick-sell credit, unopenedPacks.
Perishable intel captured: the real FIFA 17 retail pack catalogue, 41 SKUs with Origin
offer ids, recovered from the client heap as a parsed copy of data/store/storecfg.xml.
It is in no file on disk, only in a running process.
futmem/ is a standalone read-only Rust crate for this kind of work (maps, find,
strings, read). Read-only by construction: it opens /proc/<pid>/mem with File::open
and there is no code path in it that can write to another process, because a live game
session depends on that. Its own [workspace] table keeps it out of the parent
workspace. Chunked scanning overlaps by pattern_len-1 so a match spanning a chunk
boundary is still found.
utas_server.py gains FUT_PORT/FUT_LOG so a throwaway instance can be started without
bouncing the one the live client is using. Defaults unchanged (8099, /tmp/utas_server.log).
Noted for the record: this edit came from a research agent that had been told not to
touch server code. It is benign and useful, but it was out of scope.
Not committed: the doc proposes ENDPOINT_MAP.md changes as pasteable text rather than
applying them, and every proposed server change defaults off per the house rule.
Nothing in this commit changes a response the client sees.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Yesterday's settings-gate plan asserted that IS_FRIENDLY_SEASON_ENABLED and
IS_DRAFT_MODE_ENABLED "have never been set to true by anything, on any run", and
proposed spending a launch on that premise. Measured against the running client,
both read 1, and so does packOpeningAnimationEnabled, while /settings has only ever
been answered {"configs": []}.
disp 0x1fd3a (friendlySeasonsEnabled) value = 1
disp 0x1fd3d (enableDraftMode) value = 1
disp 0x1fd45 (packOpeningAnimationEnabled) value = 1
Reproduced on two separate launches and two different pids.
Where the reasoning went wrong: the finding that FUN_18011dc50 is the only writer of
those bytes, and that the FutDataManagerImpl constructor never touches them, was
correct. The inference was not. The applier runs whether or not the configs array has
content, and the struct it is handed defaults these fields to 1, so the bytes were
being written all along. "Nothing populates the array" was treated as "nothing writes
the byte". Only the first of those was ever established.
Seasons therefore does not refuse because its gate byte is false. Its gate byte is
true. That diagnosis restarts, and the live test in section 3 should not be run as
written. The doc keeps the wrong turn on the record rather than quietly deleting it.
tools/gate_byte_probe.py makes this repeatable instead of a one-off. It is read-only
(O_RDONLY + pread), resolves the pid by comm, re-derives the CardsDLL slide from
/proc/<pid>/maps rather than caching it across launches, proves the slide against the
FNV prologue at 0x180180d00 read from the on-disk PE before trusting any address, and
decodes each gate displacement out of its accessor stub (0f b6 81 <disp32>) rather
than reading it from a table. Needs the client at the FUT hub, since CardsDLL loads
only then.
Also carries the two /settings changes that were pending from before: the mode
defaults to `off` (the live-proven baseline, since nothing here has faced the game)
and the transfer-pile probe is 77 rather than 100, because 100 is a stock-looking
number that would prove nothing if it showed up in game.
Live: 439 contract checks pass. check_settings_flags.py passes in all four modes.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
ENDPOINT_MAP said this class reads one key, `configs`, and that was true and
useless. What it missed is what happens after each element closes: the client
feeds the STRING VALUE of `type` back through the atom hasher and switches on
the result, 42 arms wide. A flag is a row, not a key, and the client hashes our
string itself.
Followed it to the end. FUN_18011dc50 is the only writer of the IS_* UI gate
bytes inside FutDataManagerImpl, every line is `byte = (field == 1)`, and the
constructor never touches those bytes. So a flag nobody sends is a gate nobody
opens. friendlySeasonsEnabled and enableDraftMode have never been sent by
anything, which is a mechanism for Seasons refusing while making zero requests
to any of the four servers.
The store is the control that makes this readable: IS_STORE_ENABLED is the same
kind of byte and its screen works, because storeEnabled and friends already
ship through the Blaze config store. That list has no seasons or draft flag.
Ship the gates behind FUT_SETTINGS (off/keep/gates, default gates), and
re-assert the working store flags in the same array on purpose: once a
populated array makes the applier run, it writes EVERY gate byte, so omitting
them could switch off a screen that works today.
maximumTradePileSize=100 rides along as a positive control, because a boolean
that changes nothing cannot distinguish "the flag did not help" from "the array
never reached the consumer".
check_settings_flags.py asserts each shipped name against the atom table AND
the recovered switch, since a misnamed flag is silently inert and looks exactly
like a failed fix. enableSquadBuildingSetsFeature is the reason both checks are
needed: a real atom with no arm here.
Live: 439 contract checks pass, market unit suite passes.
Not yet tested in game.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Researched rather than guessed, after a guessed field crashed the client.
VERIFIED: FUN_1800d8330 returns cardtype 9 for exactly 0x1e, 0x1f, 0x91..0x96,
0xe7..0xe9, 0xec; fcc_misccards' cardsubtype 231 anchors the 0xe7 block to misc, so
badges/kits/stadia/balls/logos live in 0x1e, 0x1f and 0x91..0x96.
VERIFIED, and it answers a question nobody had asked: the itemState enum at
0x180229d20 is WAITING_FOR_GAME, inGame, forSale, offered, activeBadge, activeHomeKit,
activeAwayKit, activeBall, activeStadium, active. An EQUIPPED club item is the same
item with itemState set, not a different subtype. 'free' is right for owned-but-not-
equipped, which is what we already send.
VERIFIED: club items have no category group table (consumables and staff both do), and
the route is club?type= with SINGULAR names, observed live.
STILL UNKNOWN and labelled so: which subtype means which family. Not in any of the 149
dumped tables, no group table, and cardtype 9 has no merge arm so a wrong value cannot
announce itself.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
The game hung and then crashed on the first equippables fetch. My fault twice over.
CAUSE, primary. I copied teamid, leagueid and value straight out of the fcc row as
extras. `value` appears elsewhere as an OBJECT member (displayGroup {"value": ...}),
and a scalar where an object is expected is the type-desync busy loop at 0x1801c7f1a
-- which presents exactly as "the game is taking its time" and then dies. Omission is
safe; an unestablished field is not. That is this project's own rule and I broke it
for three fields that were not needed to draw a card. All three are gone.
CAUSE, contributing. The last request before the crash was type=equippables&count=11
and we answered with 30 items spanning FIVE unverified cardsubtypeids at once: the
widest possible blast radius for a wrong shape, and it tells you nothing about which
subtype was wrong. equippables now answers [] until the subtypes are confirmed one
family at a time, and shelf() takes a families= filter so a test can serve exactly one.
Adds FUT_CLUBITEMS=probe:<family>, which serves one item per candidate subtype for a
single family, so the screen names the correct subtype instead of me guessing a third
time. Eight items, one family, one question.
The flag already defaulted off, so a plain restart cannot serve any of this.
439 + 414 checks green.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
Arming the counters made the client name the route within seconds, exactly as it did
for consumables:
club?type=equippables&count=11
club?type=stadium&count=200
club?type=ball&count=200
So it IS club?type= for this family, with SINGULAR names -- stadium and ball, not
stadia and balls -- and equippables as the combined club-customisation view. Those
requests were being answered from STORE.items(), which holds no club items because
the shelf is synthetic, so they correctly returned empty and looked like nothing was
happening.
Now serves the shelf: stadium 4, ball 6, badge 8, kit 8, equippables 30 (all four
combined), with player untouched at 205.
The cardsubtypeid values remain UNVERIFIED. cardtype 9 has no arm in the merge, so a
wrong subtype cannot announce itself; whether these render is now a live question and
the next screenshot answers it.
439 + 414 checks green.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
Correcting an overstatement first: I said every card family works. Balls, stadia,
badges and kits do not, and this is the start of that, not the finish.
CLUB ITEMS (FUT_CLUBITEMS=1, default off). tools/fut_clubitems.py builds shelves from
the game's own tables: fcc_balls 42, fcc_stadium 78, fcc_badgecards 656,
fcc_kitcards 1482, fcc_leaguelogos 44, each with its real carddbid AND its real
cardassetid. Counts are wired into the club stat set, replacing the honest zeros
rather than appending a second row per id (the deserializer does
store[ctx][statId] = value, so two rows for one id is a coin toss).
Counts FIRST and on purpose. The consumables round proved the count gates the fetch:
the client does not ask for an item list until club/stats reports a non-zero count,
and the CLUB tab reads 0x1e balls, 0x28 kits, 0x14 stadia. Arming the counters is what
makes the client name the item route, which is the one thing static reading has never
produced for this family -- cardtype 9 has NO arm in the merge, so nothing about it is
discoverable from the card DB.
What is deliberately NOT guessed: which cardsubtypeid means ball versus stadium. The
eight values that reach cardtype 9 are {30,31,145..150} and the assignment appears in
none of the 149 dumped tables. probe_shelf() serves one item per candidate so the
screen can say which is which, rather than shipping a guess into someone's club.
PACKS (FUT_PACK_MIX=1, default on). A pack is no longer eleven footballers: roughly a
quarter of each pack is now consumables and occasionally staff, as a ratio so it
scales from a 5-card bronze to an 11-card premium. Club items are excluded until their
subtypes are verified -- a pack is the worst place to discover a wrong subtype,
because the card lands in the save and has to be cleaned out by hand.
Also fixes the art id AT THE SOURCE. fut_consumables now sets cardassetid from the
fcc_ tables, so a pack-granted consumable renders correctly too. The earlier fix only
corrected the copy served by the club route, which meant packs would have dealt cards
with the green NOT FOUND box again.
Pack tests run against a COPY of the profile, after an earlier test persisted 15 cards
into the live save and had to be undone by hand.
439 + 414 checks green.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
Real card art, tier colours and the GK glove icon all draw once cardassetid carries
the art id. Records both refuted hypotheses with the evidence that killed them, since
each was plausible and someone will reach for them again.
The generalisable trap: an fcc_ row has BOTH carddbid and cardassetid, they are not
interchangeable, and _item copies resourceId into cardassetid -- right for players,
wrong for every other family. Club items will hit it next: balls 37, kits 35,
stadium 36, badges 39, league logos 40.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
A player photographed a green tag under every consumable card. It is not a status
label at all: external/ion_fut/artAssets/.../notfound.swf is the client's placeholder
for an art asset it could not resolve, so the card was drawing 'no such artwork'.
The cause is two id columns that look interchangeable and are not. In the game's own
fcc_ tables a consumable row carries BOTH carddbid (5003001) and cardassetid (3), and
the art is keyed on the small one. fut_store._item copies resourceId into cardassetid,
which is correct for players and wrong for every other family, so the client looked up
art id 5003001, found nothing, and fell back.
Now mapped from data/tables/fcc_*.json, dumped read-only from the client's own
database, so these are the game's ids rather than a guess: training 3, contract 7,
healing 10, misc 45. The same table also gives balls 37, kits 35, stadium 36,
badges 39, league logos 40, which is what the club-item family will need.
Two earlier guesses at this badge were wrong and are recorded as such: it is not the
untradeable flag (changing it left the tag untouched, and the record showed the flag
had flipped) and not a loc-string failure.
439 + 414 checks green.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
A player pointed at a green tag under every consumable card. It is ours, not the
client's: the deserializer sets a flag from (untradeableCount < count), we set
untradeableCount == count, so every stack read as fully untradeable and drew the
badge. UNTRADEABLE_COUNT and UNTRADABLE_COUNT are UI state keys in .rdata.
In FIFA 17 a pack-opened consumable is normally tradeable, so this was our own data
showing through rather than a rendering fault. Now untradeableCount is 0 and the
served copy carries untradeable false. FUT_CONSUM_UNTRADEABLE=1 restores the old
behaviour.
TODO/CONFIRM: the exact badge text was not read, only its source. If the tag survives
this change it is a different label and the untradeable theory is wrong.
439 + 414 checks green.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
Rendering with real artwork, quantity badges and +5/+10/+15 amounts, so atom 0x1b
reaches record+0xbf. The client's own dialog names the class we resolved: 'Search
Type: Consumables Search'.
Records the three things that each had to be right and each failed silently with a
200: the count gates the fetch, the route is club/consumables/<category> (a /club
PREFIX, so it was being answered with the player list), and the element is a five-atom
stack wrapper whose 0x16a member is the only thing that carries the item.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
The route was right and the body was wrong. We served bare items, the client took the
response and inserted NOTHING (the live card map held only the 11 squad players), and
the screen stayed empty with no error logged anywhere. A 200 with a well-formed body
that the consumer silently discards is the worst failure shape there is, and it is the
third time this project has hit it.
FutConsumablesSearchServerResponse resolved from the RS4 literal 0x1802222f8:
factory 0x180130a10, vtable 0x180222200, deserializer +0x08 = 0x180130d10 (6873
chars). It takes itemData(0x16b) at the root like the club list, but its ELEMENT is
not an item. It is a five-atom wrapper and exactly one of the five carries the item:
0xbc count int
0xd7 discardValue int
0x16a item -> FUN_18013fe00, the item parser itself
0x287 resourceId int
0x362 untradeableCount int
Everything else goes to the value-skip handler, which is precisely why a bare item was
accepted and did nothing. It also explains the UI: FUT draws consumables as one stack
with a quantity, not as N cards, and the wrapper is that quantity.
Identical consumables are now collapsed by resourceId and counted.
439 + 414 checks green.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
The counter WAS the gate, and fixing it produced the request within seconds:
11:23:36 GET /ut/game/fifa17/club/consumables/training
11:23:40 GET /ut/game/fifa17/club/consumables/contracts
Neither is club?type=, and neither is the /consumables/%s template we had been
hunting in the binary. The client asks here, and it asks ONLY once
club/stats/consumables reports a non-zero count. Two rounds of item-shape work went
unrequested for want of a counter.
Worse, the path is a /club prefix, so it fell through to the generic route: the
consumables screen was answered with the 194-card PLAYER list. It asked for training
cards and got Cristiano Ronaldo.
Now routed above the generic /club, serving the shelf filtered by category. The
segment names come from the UI group table at 0x180203260; training and contracts are
CONFIRMED on the wire, the other five are from that table and matched case
insensitively, with singular "contract" accepted because the client has used both.
An unknown segment serves the whole shelf and logs loudly rather than showing an empty
screen, since a new spelling is a wire fact worth catching.
Per-category counts match the on-screen panel exactly: training 42, contracts 13,
fitness 6, healing 21, position 20, playStyle 24, managerLeague 0. That correspondence
is what makes an empty list distinguishable from a wrong mapping.
439 + 414 checks green.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
The tab was empty because we answered the wrong question. The client asks
GET club/stats/consumables 41 times a session; we replied with the PLAYER stat set
(players 205, playersGold 189 ...), which that panel does not read. Confirmed on
screen: seven categories present and selectable, every one reading 0.
Now appends 14 consumables* rows counted from the SHELF. The shelf, not STORE.items():
the consumables we serve are a synthetic overlay never granted into the save, so
counting the store gives fourteen zeros, which on screen is byte-identical to failure
and would have made the experiment unreadable. Counts match the independently derived
expectation exactly: 126 total, 21 healing, 7 player contracts, 21 player training,
3 player fitness, 20 position, 21 GK training, 6 manager contracts, 19 playstyle.
Safe by construction: the vocabulary is an ATOM switch (FUN_18012fd40, 40 arms,
default return 0), so an unrecognised name is inert rather than fatal, and the rows
are APPENDED -- the player, nation and league rows that drive the working screens are
untouched. Zero rows unless FUT_CONSUMABLES is armed.
Default ON because answering the consumables panel with player counts is wrong by
inspection rather than a judgement call. FUT_CONSUM_STATS=0 reverts.
439 + 414 checks green.
The open question this sets up: whether a non-zero count makes the client request an
item list at all. If it does, the log names the route and /consumables/%s is settled
for free. If the numbers move and no request follows, the panel renders from counts
alone and the 126-item shelf was never needed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
Round 3, 10 agents. The headline is measured, not inferred: GET club/stats/consumables
is requested 41 times per session by the real client (ProtoHttp), and _club_stat_set()
answers it with the PLAYER stat set. The panel reads 14 consumables* names that we
have never sent, so it is told '205 players' when it asked how many contracts the club
owns, and it has nothing to show.
That also explains why last round's 126-item consumable shelf was never requested. It
serves type=contract|training|healing|development and an UNTYPED /club with no
team=/league=, and all 9 of the client's untyped requests this session carry team=. The
only +126 item(s) line in the whole log came from one of our own probes.
Vocabulary recovered: the 14 consumables* rows plus badgeDBid 0x2e, kitsHome 0x29,
kitsAway 0x2a, leagueLogos 0x2f, trophiesSeasonOnline 0x38.
Other measured surfaces the client asks for and we fob off: GET /settings 11x answered
with an empty config array (a 40-flag feature gate, the biggest untouched lever in the
project), leaderboards/options 5x with {}, user/accountinfo 4x with {}.
club/stats/staff is a DIFFERENT class (FutStaffBonus); the staff counts come from the
Stats2 store, which is why the staff screen worked while we answered {}.
Refuted: ENDPOINT_MAP's claim that objectives have no route. FUN_180151610 builds
<base>/objective/%d/reward and FUN_180147780 builds .../complete.
New modules only. utas_server.py is deliberately untouched: whether to wire the counts
depends on a free observation the human can make on the client that is already running,
and spending a restart before that is what this round exists to avoid.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
34 staff cards, zero DB Error. Every coach id hit its table first time, which is the
payoff from enumerating the game's own database instead of sweeping for ids: the loud
miss-fill exists and never fired.
10 of 10 managers resolved with historically correct nations and leagues.
RESOLVED, previously TODO/CONFIRM: the manager card template paints record+0xde
(nation) and record+0xe0 (league). Luis Enrique draws the Spain flag and 'LaLiga
Santander'; the Premier League managers draw their flag and 'ENG 1'. The merge never
writes either field, so nothing but our own JSON could have supplied them.
Corrected: negotiation at +0xe3 is NOT on the card front, which shows CONTRACT 7
there. I had told the user to look for it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
Deleting cards from someone's club is their call, not the tool's. The nine
unrepairable blanks are now KEPT unless --delete-dead is passed. A blank card is ugly,
not harmful, and the 175 stale cards were never the deletion candidates anyway: they
are real players wearing old invented numbers and they get repaired in place.
Also records the build round's synthesis as docs/plan-2026-08-05-families.md.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
os.environ.get returns the STRING "0", which is truthy, so anyone typing
FUT_CONSUMABLES=0 to turn the family off would have turned it on. "", "0", "off",
"no" and "false" now all mean off; any other value is the mode string ("1", "all").
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
Wires the three families into club_route as a synthetic OVERLAY. Nothing changes by
default: with all three flags unset every route returns byte-identical bodies, checked
against the live 194-item save (club 194, type=player 194, league=53 drill-down 67,
hub clubPlayers 194, every staff stat 0).
FUT_CONSUMABLES=1 serve on type=contract|training|healing|development
FUT_CONSUMABLES=all ... and on an untyped club fetch with no team=/league=
FUT_COACHES=1 serve on type=headcoach|gkcoach|physio|fitnesscoach|staff
FUT_COACHES=all ... and on type=manager, the one staff request ever observed
FUT_MANAGERS=1 serve on type=manager|staff
WHY AN OVERLAY AND NOT A GRANT. Clearing the flag restores the real club exactly on
the next fetch, with no un-granting and no edit to a save a live client is holding
open. And Store.add_items() uses setdefault("id", ...), so items arriving with an id
already set do NOT advance nextItemId -- granting these would eventually collide two
id spaces. The four overlay bases (9.4e8 consumables, 9.5e8 coaches, 9.6e8 managers,
9.1e8 probe) are asserted disjoint from each other, from the save's 1e8 and from the
sweep's 9e8. The cost is that overlay cards cannot be quick-sold or moved.
WHY EACH FLAG IS TWO-VALUED. The tab-to-?type= binding is UNOBSERVED -- only
type=player, type=manager and type=custom have ever come from this client, and none of
the nine other arm names in FUN_18012ec50 has. So every club fetch now LOGS the arm it
was asked for while any flag is set. That is what makes an empty tab actionable: it
says whether the request reached us and under which name, instead of nothing.
THE MIRROR FILTER, and it is not optional. club_route applied its cardsubtypeid filter
only when `kind and kind not in ("player","custom")`. For type=player, for type=custom
(the by-league / by-team drill-downs) and for an untyped fetch it filtered NOTHING, so
the moment the club held a non-player item it would be served straight into the
players tab and the drill-downs -- and a manager carries nation, leagueId and teamid,
so he would have appeared as a footballer in exactly the MY CLUB rows that were only
just made non-zero. The player branch now filters cardsubtypeid in (0,1,2,3). Provable
no-op today: all 194 items in the save are cardsubtypeid 0.
Same hardening for the three counters that keyed on itemType == "player" (hub
clubPlayers, _club_stat_context, _club_stat_set) -> _is_player(), i.e. the CLIENT's own
definition from FUN_1800d8330. itemType is INERT on the wire (atom 0x173 is parsed into
a stack std::string in FUN_18013fe00 and freed; it never reaches the record), so keying
our own screens on it made their correctness depend on a field the client ignores.
FREE SECOND ORACLE: the five staff sub-type counters (0xb..0xf) are now computed from
the overlay. FUN_180094ce0's STAFF_EMPLOYED row is the +0x800 SUM over those five, so
the parent id 0xa alone could never move it. That number moves without the merge being
involved at all, which keeps "our club reports N staff" separable from "the client
resolved the card".
item_def() also learns consumables (gated): answering a consumable resourceId with
cardsubtypeid 0 makes it cardtype 0 -- no merge arm, no miss-fill, i.e. plausible
garbage -- and it carried the same hardcoded rareflag 1 as fut_store._item().
tools/test_card_families.py: 414 offline checks over the item BUILDERS. Deliberately a
separate suite -- test_fut_contract.py talks to a live server over HTTP and imports
nothing from the server's own code, which is what lets it certify a future non-Python
implementation; these must run against the working tree without restarting anything.
Each guard was mutation-checked: reverting the 219 fix, or letting coach_item accept a
miss-fill assetid, or letting consumable_item accept a dead zone, each fails the suite.
Suites after: test_fut_contract 439/0, test_match_rewards 61/0, test_card_families
414/0. utas_server was NOT restarted; this lands on disk only.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
managercards: 417 rows, carddbid 1000001..1001552, assetid == carddbid on all 417,
value 57..88 (120 rows at exactly 57), rare 282/135 and NOT a rating threshold.
talkrating and formationid are 0 on ALL 417 rows -- both columns are dead in FIFA 17.
MEASURED from data/tables/, rowcount == rows_emitted.
Names come out after all, and by the CLIENT's own rule rather than an inference:
FUN_1801bb060 (879 bytes, read in full) does SELECT firstname,surname,... FROM manager
WHERE managerid == (*(u32*)(rec+0x18) & 0xffffff) - 1000000. Joining data/tables/
manager.json on that rule gives 1000509 Luis Enrique 88, 1000089 Wenger 86, 1000417
Guardiola 87, 1000414 Klopp 84 -- the ratings match the men, which a shifted column
could not produce. This supersedes the note that we get ids and not names: true of
managercards, false once you join `manager`.
Fixed here: `manager` has 747 rows but 746 distinct managerids -- managerid 107 is
duplicated with an empty-name row, and a plain dict comprehension kept the wrong one,
silently giving carddbid 1000107 a blank name and teamid 1357 instead of Slutskiy and
315. 296 -> 297 usable names.
THE KEY FACT, verified at instruction level: the parser routes the JSON `nation` to a
DIFFERENT record offset for a manager. At 0x180140e0b FUN_1800d8330's result is DEC'd
twice -- cardtype 1 stores nation to rec+0x148, cardtype 2 to rec+0xde, everything
else DISCARDS it. leagueId (atom 0x18a) lands unconditionally at rec+0xe0. The manager
merge FUN_1801356c0 (452 bytes, 1,587 chars, read in full) writes only firstname,
lastname, assetid, rating, talkrating, negotiation and rare -- it never touches
rec+0xde/+0xe0. So for a manager, WE are the only source of nation and league, and
they are read: FUN_1801a8580/FUN_1801a8540 feed FUN_1800e5940 ("ManagerCardBio"),
which publishes NATIONALITY, NATIONALITY_ASSET_ID and LEAGUE_ID.
That merge has NO else-branch, so unlike a coach a wrong manager id is COMPLETELY
SILENT. resourceId must equal carddbid exactly -- the manager arm reads the key raw,
with no & 0xffffff.
Also corrected against the design round's own draft: talkrating and negotiation are
NOT unread. FUN_1800e5940 publishes them as ATTRIB_TEAM_TALKS (rec+0xe2) and
ATTRIB_CONTRACT_NEGOTIATION (rec+0xe3). They come from the DB, not from us, and since
talkrating is 0 on all 417 rows TEAM TALKS reads 0 on every manager card in the game.
Not wired into the server in this commit.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
headcoachcards 124 rows 2000004..2000328, gkcoachcards 121 rows 9000001..9000324,
physiocards 51 rows 4000002..4000259, fitnesscoachcards 115 rows 3000019..3000328.
All MEASURED from data/tables/, dumped read-only from the running client; rowcount ==
rows_emitted == len(rows) on all four, which is what makes "this id is absent" a claim
about a complete dump rather than about a truncated one. assetid == carddbid on every
row; every id fits in 24 bits.
WHY COACHES ARE THE CHEAPEST FAMILY TO TEST. Their four arms of FUN_180141660 (2,129
bytes, 214-line decompile read to its closing `return`) are the only merges in the
game that label their own failure: on rowcount < 1 each writes firstname = lastname =
"DB Error", rec+0xb4 = 0x32, rec+0x58 = 1 and a TABLE-UNIQUE assetid -- head 2000148,
fitness 3000259, physio 4000146, gkcoach 9000258. Two independent facts make that a
one-glance oracle, both verified by exhaustive scan of all 411 rows: no row in any of
the four tables has value == 50, and no fitnesscoach row is (fieldpos 1, posbonus 7,
amount 1).
CORRECTION to docs/plan-2026-08-04-card-families.md: the miss-fill is NOT uniform.
Only head coach and GK coach write 0xf into the attribute array at rec+0x98. Physio
writes 0xf into a BYTE at rec+0xdd; fitness coach writes no 0xf at all -- rec+0xde =
0x107 and rec+0xdd = 1. So card_identity_probe's attrs column means something
different per family, and its F_NAME_KNOWN=0xdd string read sits directly on top of
physio's, fitness coach's and the manager's raw stat bytes. Use coach_probe.py.
The key is RAW: all four staff branches pass *(u32*)(rec+0x18) unmasked into
`WHERE carddbid == ?`. Players are the only family that masks with & 0xffffff, so a
version byte in the top octet breaks every staff lookup -- silently on a manager,
loudly on a coach.
WHAT WE SEND: id, resourceId, cardsubtypeid, itemType, contract, itemState, owners,
untradeable. Nothing else. rating/rareflag/assetId are overwritten by the merge;
nation/leagueId/teamid would be INVENTED, because none of the four tables has such a
column; preferredPosition (rec+0x146) and attributeList (rec+0x98..) SURVIVE the merge
and are read by the generic view-model FUN_1800d7920, so sending them would hang a
position label and six attribute numbers on a coach. Omission is safe; a scalar where
an object is expected is not.
The starter shelf is one card per (tier, rare) combination per family -- 24 cards --
with two exclusions: the four miss-fill assetids (three of which are REAL rows, so a
hit and a miss would look identical on those cards), and any row whose own stat write
is byte-identical to its family's miss-fill (head/GK attribute 0 amount 15).
tier() is the binary's own tail, not our convention: the shared exit of FUN_180141660
writes rec+0x54 = 3 if rating >= 0x4b else 2 - (rating < 0x41), for every arm
including the miss arms.
Also lands the design round's read-only probe tooling: coach_probe.py (grades a live
record HIT/MISS/WRONG-BRANCH/NO-MERGE against the on-disk rows) and coach_window.py
(builds a mixed-control window; fires nothing).
Not wired into the server in this commit.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
144 live cardtype-6 subtypes and 28 dead zones, all derived from the binary rather
than guessed, plus EA's own authored variants out of the dumped fcc_* tables.
MEASURED (decompiles read to their closing brace, lengths stated):
FUN_1800d8330 (714 chars) cardsubtypeid -> cardtype. The cardtype-6 space is
{51..136} u {201..220} u {250..273} u {300..341} = 172.
FUN_18013f4d0 (8,354 chars) subtype -> category(rec+0xb8), sub-sel(rec+0xbc i16),
amount(rec+0xbf i8), single(rec+0xc0). Two callees, a
range clamp and an enum map; no DB handle is touched,
which is why a consumable has no identity to look up.
FUN_1801bfac0 (42,813 chars) category -> FUT_CONSUMABLE_* string + a HARDCODED
5000xxx artwork constant. resourceId never reaches the
screen for a consumable.
fcc_trainingcards 143 rows / fcc_healingcards 27 / fcc_contractcards 13, each with
rowcount == rows_emitted == len(rows), so absences below are from a COMPLETE dump.
Two things the family will not forgive, both enforced in the builder rather than
documented and hoped for:
* `amount` (atom 0x1b) is MANDATORY for categories 0, 4, 5, 9, 10. The parser
initialises its temp to 0xffffffffffffffff, so omitting it stamps (byte)-1, and
the accessors FUN_1801a8040/FUN_1801a8060 are `(int)*(char *)` -- SIGNED. The
card reads "-1", not 0. consumable_item() raises instead.
* A DEAD-ZONE subtype does not self-label. It falls to the bottom default of
FUN_18013f4d0 and renders as an ordinary Squad Training (Pace) card with amount
0. There is no "DB Error" analogue here, so every subtype we ship comes from
data/consumables.json and the builder refuses the other 28.
Two corrections to the generated data, both from re-reading FUN_1801bfac0 case 5 and
case 0 rather than from the category table:
* subtype 220 is named FUT_CONSUMABLE_NAME_SQUADTRAINING, not ..._PLAYERFITNESS.
0xdc == 220 is the FIRST half of the squad-fitness test, so 220 always takes that
branch, and there is no ..._SQUADFITNESS string in the binary at all.
* all 28 dead zones are SQUADTRAINING, not PLAYERTRAINING: case 0 tests
`subtype - 0x33 < 7` then `subtype - 0x3d < 7` and no dead zone satisfies either.
Exactly 29 of 172 rows changed; nothing else moved.
INFERRED, and flagged as such in the module: the ?type= grouping. The vocabulary is
certain (FUN_18012ec50 arms healing=23, contract=24, training=25, development=6), but
the tab-to-arm binding has NEVER been observed -- only type=player, type=manager and
type=custom have ever come from this client.
Three families deliberately NOT shipped: manager_formation_mod (71-86) has zero rows
in the 143-row table AND FUN_1801bfac0 case 6 calls FUN_1801a0100 on the formations
result without the rowcount guard its twin case 7 has -- a crash candidate;
formation_mod (121-136) has artwork -1; manager_league (300-341) renders literally
"ML: %d" from a raw number and one shipped amount (2118) is in no league table.
Not wired into the server in this commit.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
fut_store._item() hardcoded "rareflag": 1 on every item it builds. That is inert
for players and for staff, and CORRUPTING for exactly one consumable subtype.
MEASURED, in the binary: FUN_1801bfac0 case 5 (consumable category 5, fitness)
takes the squad-fitness branch when
(cardsubtypeid == 0xdc) || FUN_1801a88c0(rec)
and FUN_1801a88c0 is exactly `*(int *)(rec + 0x58) == 1`. rec+0x58 is the rareflag
atom 0x271 (FUN_18013fe00 case 0x271 -> uStack_130; the frame arithmetic is
independently pinned by local_138 -> rec+0x50 and local_13c -> rec+0x4c, the two
offsets card_identity_probe has been reading live for days). FUN_180141660 does not
overwrite rec+0x58 for cardtype 6 -- cases 6/7/8/9 fall to the shared tail, which
writes only rec+0x54 -- so a rareflag we send survives all the way to the render.
Result: subtype 219 with rareflag 1 draws FUT_CONSUMABLE_NAME_SQUADTRAINING with
artwork 5000011 instead of Player Fitness with 5000010, and forces the
single-target count at param_5+0x1bc to 0. Silent. It would have corrupted the
first fitness card we ever served.
The guard is `0 if cardsubtypeid == 219 else rareflag`, added with two new KEYWORD
params. Every existing call site (fut_store.py:74/:357, utas_server.py:1404/:2136)
passes 8 positional args, so both take their defaults and the player dict is
byte-identical -- key order included, asserted in tools/test_card_families.py.
Scope correction to the round's own notes: rec+0x58 is read TWICE in that
42,813-char render function, not once. FUN_1801a88c0 is the category-5 read, but
line 108 reads it directly into param_5+0x1f0 (the rare/backing art) for EVERY
cardtype, before the `if (param_4 == 6)`. So the guard also stops a 219 being drawn
as rare -- intended, since rare IS the squad-fitness selector.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
Audit of the live club: 194 items, 10 already correct, 175 stale, 9 dead.
STALE means a real FIFA 17 player carrying the old invented fields: attributes
computed from the rating, and in some cases a wrong club or nation (Kroos was stored
at Bayern and is at Real Madrid; Alaba was nation 40 and is Austria 4). Those are
REPAIRED in place from data/pool.json rather than deleted. Deleting them would throw
away 90 percent of the club for no reason: name, face and badge are already right and
only the numbers are wrong, and the item id does not change so squad slots survive.
DEAD means the playerid is not in the roster at all, so the client misses and stamps
its generic blank. Nothing to repair, so those are removed. A dead card referenced by
a saved squad is kept rather than breaking the slot.
The nation and team mappings were spot-checked against the game's own nations and
teams tables before trusting them across 175 cards: 4 Austria, 21 Germany, 38
Portugal, 60 Uruguay, 243 Real Madrid, 5 Chelsea.
MUST RUN WITH utas_server STOPPED. The server keeps the profile in memory and
rewrites it on its own schedule, so an edit underneath a running server is clobbered
by the next save. That is why the nine dead cards removed on 2026-08-04 were back a
few hours later: the removal was correct and the running server undid it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
Reported live: a Cristiano Ronaldo card appearing under Chelsea and under Arsenal.
The data was right and so was the client. teamid 243 really is Real Madrid in the
game's own teams table, and all eight Ronaldo cards in the live CardsDb map read
teamid 243. The fault was ours: club_route parsed only ?type= and ignored ?team= and
?league=, so clicking a club or a league in the club panel was answered with the
ENTIRE 194-item club. Every drill-down therefore contained every player.
Note which half was already correct: the club/stats COUNTS were fixed yesterday and
were right (England 11, Premier League 17). It was only the item list behind them that
was unfiltered, which is why this looked like a data bug and was not one.
Now: team=5 gives 26 items, team=243 gives 20, league=13 gives 22, and the unfiltered
club list is untouched at 194. 439 checks green.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
The pool now comes from the game's OWN resident database, not from a rating index
plus guesses. tools/db_dump.py walked the client's self-describing table catalog
read-only and wrote data/tables/ (149 tables, 55MB); tools/build_player_facts.py
turned it into data/player_facts.json; data/pool.json is the compact form fut_cards
loads.
MEASURED, per player: position (players.preferredposition1), nationality, teamid,
leagueid (via leagueteamlinks), and the six card attributes.
The six attributes are NOT columns -- they are a weighted sum of the 29 base
attributes, and the weights are read out of the game's own playerattributesmapping
table rather than from published formulas. Checked against real FIFA 17 cards:
Messi 89/90/86/96/26/61 and Ibrahimovic 72/90/81/85/31/86 are EXACT, Suarez is one
off on physical, Ronaldo within two on pace and shooting. Keepers come out directly
from the gk* columns.
What this fixes on screen: Kaka was a CDM, Bale a CM, Suarez a GK, and every
attribute was derived from the rating. Now Bale is RW, Boateng is a CB with 90
defending, De Gea is a GK, and a bronze pack deals real bronze players in real
positions.
REVERSAL, deliberate: nation/team/league were being sent as ZERO so the client would
fill its own values (the merge fills those three only when they arrive zero). Now
that we hold the game's own numbers there is nothing to gain from zeros, and they
actively hurt -- club-stats drill-downs bucket by the item's own nation and leagueId,
so a club full of zeros would have quietly emptied the per-nation and per-league
panels fixed the day before. Send the real values.
The old rating-index path is kept as a fallback so the pool still builds without
data/pool.json, and it now says out loud which of the two it used, because one is a
measurement and the other is a guess.
439 + 61 checks green.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
11-agent round, every investigation adversarially reviewed. The headline is that this
was never five id hunts: the client's database is resident in ordinary heap as a
self-describing catalog of bit-packed fixed-stride row arrays, walkable READ-ONLY, so
the id sets fall out at zero cost in human club visits.
MEASURED:
managercards carddbid 1000001..1001455, assetid == carddbid; two agents using two
different block locators agreed 417/417 (docs/managercards_ids.txt). This is why
the earlier sweeps of 1..5000 and 6000..8000 were silent.
staff bands: headcoach 2000004+, fitnesscoach 3000019+, physio 4000002+,
gkcoach 9000001+, corroborated by the four miss-fallback assetids hard-coded in
FUN_180141660 each landing inside its own table's decoded range.
all four coach branches write a LOUD miss-fill: firstname/lastname "DB Error",
rating 0x32, rare 1, attrs[0] 0xf, plus a table-unique assetid. Managers write
none, so a wrong manager id is silent and a wrong coach id labels itself.
consumables have NO table and NO id space: cardtype 6 has no arm in the merge,
FUN_18013f4d0's only callees are a range clamp and an enum map, and every string
is a hardcoded FUT_CONSUMABLE_* literal. A contract is three JSON keys.
the ?type= taxonomy is 29 explicit arms plus a default: badge 11, kit 12, stadium
13, ball 14, equippables 15, leaguelogos 16, misc 26. club/stats kits and
badgeDBid are PLAIN COUNTS, not ids.
fancards is a boolean column of the fixtures table; newcards is FUT atom 0x1d7.
NEITHER is a card family, so two of the five hunts never existed.
REFUTED, and worth keeping: the live table-directory walk was off by one entry
(descriptor for table T is at entry-0x20, not entry+0x08), which had mislabelled
managercards as factory_teams and shifted every column count. managercards names are
32-bit string-pool offsets and the pool was never located -- we get ids, not names,
and we do not need names because the client supplies them.
TRAPS RECORDED: rareflag=1 silently converts a Player Fitness card (219) into Squad
Fitness, and fut_store._item() hardcodes rareflag 1 on every item.
Four proposed club-item sweep windows were killed by review as invariant by
construction: with no merge arm there is no miss-fill, so every id yields a
byte-identical record and the probe cannot discriminate. That is a wasted human action
correctly caught before it cost one.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
Supersedes the parts of this document that were wrong: the CardsDb map is not empty
offline, and dbdata.dll is not the player database.
Records the merge dispatch table, the field-fill asymmetry that the pool design
depends on (send zero for what the client knows, send our own only where it knows
nothing), the three-state oracle with all three fingerprints, the 5000-item ingest
ceiling, the fact that the map is WIPED on every club fetch, and where the 17,547
player roster came from plus its independent cross-validation.
Also records the state of the other card families so the next session starts from
the manager branch writing no miss-fill, rather than rediscovering it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
The manager sweep (subtype 4, ids 1-5000) came back with 5000 records at
cardtype 2 -- confirming FUN_1800d8330(4)=2 live -- and NOTHING written: no name,
no nation, no teamid, and no miss-fill either. Our sentinel rating 7, position 25
and attributes all survived. So either manager ids are not in 1-5000 or that
branch keys off something the player branch does not.
Guessing the id space costs a club visit per guess, so 't*@lo-hi' now fans all
five non-player tables across one range in a single response: 4 managercards,
5 headcoachcards, 6 gkcoachcards, 7 physiocards, 8 fitnesscoachcards. One staff
tab load tests 1000 ids against all five.
The full table set, from the DLL's own strings: players, managercards,
headcoachcards, fitnesscoachcards, gkcoachcards, physiocards, fancards, newcards
-- plus a consumables family (contract, fitness, healing, position, training and
playstyle modifiers, formation and league mods) and club items (badges, balls,
kits) which are almost certainly NOT DB-resolved the way cards are.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
Observed live: the staff tab issues GET /club?year=2017&type=manager&count=200.
club_route ignored the parameter entirely and answered every type with the full
player list, so FUT displayed players as coaching staff.
The filter is deliberately narrow. 'player' and 'manager' are the only values the
client has ever been seen to send; 'custom' (the by-league and by-team drill-downs)
and a missing type keep exactly the behaviour that is already proven on screen,
because those drill-down counts were only just fixed and must not be disturbed. An
unrecognised type is filtered rather than answered with everything, since
answering an unknown question with the whole player list is the bug being fixed.
We own no staff cards, so type=manager is [] today -- an empty item list, the same
shape the itemData parser already accepts everywhere else.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
The merge FUN_180141660 dispatches on record+0x4c, which FUN_1800d8330 derives
from cardsubtypeid alone, and each branch queries a different table by
carddbid = record+0x18 -- the same field players use for playerid:
0..3 -> 1 players (live-proven) 5 -> 3 headcoachcards
4 -> 2 manager 8 -> 4 fitnesscoachcards
6 -> 10 gkcoachcards 7 -> 5 physiocards
9..b -> 7 unidentified absent -> 0x156 -> 0, no merge at all
So a 't<subtype>@' prefix on the sweep window probes any of them the same way
players were probed: 't5@auto:1-20000:5000'.
Only cardsubtypeid changes. itemType stays 'player' because the merge dispatches
on the subtype alone and the wire shape of a real staff item has NEVER been
observed -- across every logged session the client has only ever asked for
type=player and type=custom. Inventing a shape for an unobserved request is the
change class behind every freeze this project has had.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
Leftovers from the invented-id pool that data/roster.json replaced. The client
misses on them and stamps its generic card (rating 50, teamid 1933, nation 14,
position 2, attributes 1, blank name), which is what a blank card on screen IS.
Removed 9 from the live club (5 distinct bad ids, 169193 four times over). 188 of
194 cards were already resolving; these were the whole remainder.
Refuses to remove anything a saved squad references, backs up first, and is a dry
run unless --fire. It touches only the club pile: a card present in both club and
purchased is the known fatal desync, and deleting from one pile cannot create that.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
The old pool was 79 hand-written rows whose asset ids were mostly invented, on
the premise that the client's card map is empty offline so no id could render.
That premise was refuted by a live screenshot, and this replaces its consequence.
Source: tools/dbdata_extract.py reads FIFA's own rating-sorted index out of a
running process (0x40 stride, self-validating {begin,end,end+1} name-pointer
triple, anchored on 20801 = Ronaldo 94) -> data/roster.json. dbdata.dll was a
dead end and is documented as such: its single export getTableData is an
anti-tamper attestation routine, not a data accessor.
Cross-validated against a completely independent method. tools/sweep_collect.py
serves candidate ids as a synthetic club and reads back the identity the CLIENT
resolved through its own merge. 573 of 573 overlapping names agreed exactly, and
the single id present in one and not the other is 26501, the target of the
documented 22800..22879 Legends remap -- which is also what produced 'Alex Hunter
x80' in a sweep and had looked like a bug.
Field honesty, because half of these are real and half are not:
playerid/rating/name REAL the roster
club/nation/league REAL we send zeros and the CLIENT fills them (the merge
only fills those fields when they arrive as zero)
position PARTLY 59 from the game's own per-card cache, 17 curated
by hand, the rest synthetic but deterministic
attributes SYNTH derived from rating and position
169193 is dropped from the curated set: it was in VERIFIED_ASSET_IDS and is not a
real player. The client resolves it to the database's empty placeholder row, which
renders as 'Jamal Blackman'. Two independent methods agreed.
NOTE BEFORE PUSHING ANYWHERE PUBLIC: data/roster.json is EA's player data,
extracted from your own installation. Fine locally; think twice about publishing.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
The oracle is three-valued, and all three fingerprints are now confirmed live
against a running client rather than read out of Ghidra:
NAMED our sentinel rating 7 survives and a real name appears. The id is
real, and teamid/nation/leagueId come back FILLED by the game
because we send them as zero.
placeholder rating 7 survives but the name is 'Jamal Blackman', team 0. The
players-table row exists and is an empty slot. This is the trap:
169193 does this and it was in VERIFIED_ASSET_IDS.
MISS rating 0x32, teamid 0x78d, nation 0xe, position 2, name ' '. That
is the binary's miss-fill, byte for byte, and it is exactly the
blank card photographed in a pack today.
Scale: 5000 candidates per response ingests cleanly; 20000 was served and then
silently not ingested (the map did not change at all), so the ceiling is between
them and auto chunks default to 5000.
Auto-advance: the client PAGES the club, so one visit yields several fetches.
'auto:lo-hi:step' hands out the next chunk per fetch. Item ids derive from the
candidate's offset in the WHOLE range, not its index in the chunk, so chunks never
collide and results accumulate across fetches for a single probe at the end.
sweep_collect.py accumulates into data/players.json and rejects the placeholder
name as a matter of course. Yield in 20000-24999 was 19 real ids per 5000, which
is why auto-advance matters: the real roster clusters in 150000-240000.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
dbdata.dll is an anti-tamper decoy (getTableData returns a self-integrity blob),
so the players table only exists inside the running client. But we do not need to
unpack it: the client merges its own DB into every item we serve, keyed on
resourceId & 0xffffff, and leaves the result in a map we can already read.
So serve a RANGE of candidate playerids as a synthetic club, then read the map
back with card_identity_probe.py. One club fetch classifies the whole window.
Sends teamid/nation/leagueId as ZERO so the client fills the REAL values (the
merge only fills zeros -- confirmed live: one playerid appears twice with two
different nations, both ours). Sentinel rating 7, deliberately not 50, so the
miss-fill (rating 0x32) can never be mistaken for a surviving sentinel.
The window comes from a control FILE read per request, not just the env: a full
sweep is many windows and restarting mid-session is what produced 'error
connecting to FIFA 17 Ultimate Team' once already. Nothing is written to the save,
so clearing the file restores the real club on the next fetch.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
Adds tools/card_identity_probe.py, a read-only /proc/PID/mem walk of the CardsDb
card map that reports the identity the CLIENT resolved for every card it holds.
Why it matters: identity never comes from us. The item-parser tail registers every
parsed item into the map, and immediately before that FUN_180141660 -> FUN_180135890
queries the client's own local players table by resourceId & 0xffffff. On a hit it
fills name/face and leaves our rating/position/attributes alone; on a miss it writes
a fixed generic card (rating 0x32, teamid 0x78d, nation 0xe, position 2, attrs 1,
name ' '). That miss fingerprint is exactly the blank card photographed in a pack
today, so the chain is confirmed by live evidence and not only in Ghidra.
First live run, 11 nodes, 0 failed reads, size counter agrees with the walk:
Ronaldo/Messi/Suarez/Kroos/Hazard all resolve with real names, so every record
offset derived statically (+0x18 resourceId, +0xb4 rating, +0x94 teamid, +0x148
nation, +0x146 position, names inline at +0xb8/+0xc8/+0xdd) is correct live.
This makes card identity a pure DATA problem: serve real playerids. The probe is
the bulk oracle for finding them -- N candidate ids served, one read classifies all N.
Also flips FUT_STORE_DISPLAYGROUP to default on; it shipped off pending proof that
the key does not switch FIFA17.exe to another tile render path, and it was then run
live and the store tiles showed their real names.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
CARD_SYSTEM.md has claimed since it was written that offline the CardsDb map is EMPTY,
every lookup misses, and the view-model reads every rendered field from the resolved
record and NEVER from our item JSON. A live pack open falsifies both halves.
One bronze pack, five cards. Two rendered as real players with names, club badges and
national flags. Three rendered blank: rating 50, position RWB, every attribute 1. Our
pool contains no rating 50, no RWB and no all-ones attributes, so the blank is the
client default.
The two that resolved match our item JSON field for field:
(232517, 62, RB, nation 36, league 19, team 175, [72,44,58,60,62,61])
-> SILVA, 62 RB, Wolfsburg badge, Norway flag, 72 PAC 44 SHO 58 PAS 60 DRI 62 DEF
(235066, 60, GK, nation 34, league 31, team 48, [62,63,33,61,17,62])
-> NOWAK, 60 GK, 62/61 63/17 33/62
Those attribute numbers were invented by hand this afternoon. They cannot have come
from a database.
So: stats come from our JSON, name/badge/flag come from the client keyed by assetId,
and an assetId the client does not know collapses the WHOLE card to the blank, which
is why a bad id looks like a rendering failure rather than a lookup failure.
THE CONSEQUENCE IS THAT THE PLANNED WORK IS UNNECESSARY. This document recommended
populating the map by driving the insert, hand-building a red-black tree node, or
patching the resolve miss path, all of which write to a live process. None of it is
needed. The rule is: use asset ids that exist in the client database. fut_cards.py has
18 verified ids and 61 structural placeholders, and the placeholders are the blanks.
The remaining work is a DATA problem, not a code-injection problem.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
LIVE 2026-08-04: the ENGLAND tile read 0 while drilling into it showed Premier League
17. Same bug as before, one level up: the LEAGUE buckets were keyed and the NATION
buckets were not.
The eight-row MY CLUB panel is FUN_180094ce0 (not FUN_180043b90, which is a different
provider using a different string family), and it computes:
PLAYERS_EMPLOYED = +0x7f8(nationId, 4) + (nationId, 3) + (nationId, 2)
STAFF_EMPLOYED = +0x800 over 0xb, 0xc, 0xd, 0xe, 0xf
TROPHIES_WON = +0x800 over 0x33 .. 0x38
STADIA_OWNED = +0x800(0x14) BALLS_EARNED = +0x800(0x1e)
Two consequences:
1. The no-id modes (year / consumables / club / newcards), which is what the client
fires on entering MY CLUB, now carry PER-NATION buckets keyed by nation id. The
three screens are consistent at last:
no id -> nation buckets (the tab strip and the eight-row panel)
country/<id> -> league buckets (the leagues in that nation)
league/<id> -> team buckets (the teams in that league)
2. STAFF_EMPLOYED and TROPHIES_WON are SUMS OF SUB-TYPES. Sending staff(0xa) or
trophies(0x32) alone can never move those rows, whatever their value. The eleven
sub-type rows are now emitted: staffManager/HeadCoach/GKCoach/Physio/FitnessCoach
and trophiesOffline/Online/FeaturedOffline/FeaturedOnline/SeasonOffline. All zero
today because the club owns no staff and has won nothing, but the mapping is what
matters when it does.
All eleven new type strings verified against docs/fut_atoms.tsv, 0 mismatches.
Live: /club/stats/year now returns 117 rows across 16 nation buckets, England
(nation 14) summing to 11 players. 439 + 61 checks green, zero tracebacks.
This is the third correction to this one endpoint today. The pattern in all three is
identical and worth stating once more: the parser accepts anything, and only the
CONSUMER tells you which bucket and which type ids it reads. Every time I reasoned
about the body instead of reading the reader, I shipped a wrong one.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
The tile's big number is `clubPlayers` (atom 0x90) in the body of GET ut/%s/hub, a
route we have answered with {} for the life of the project.
The chain, re-derived independently by two agents (one via Ghidra, one via raw PE plus
capstone with no decompiler) and checked by two reviewers:
clubPlayers(0x90) --INT getter 0x1801c79d0--> clamp FUN_1800d7b30 (<=0 becomes 0)
-> R+0x3c, where R = FUT data-manager slot +0x1f8 (FUN_18011a810 is literally
`lea rax,[rcx+0x1fd70]; ret`)
-> read by FUN_1800b0250, published as TEXT0 of TILE_ID 0x210
-> captions FUT_GH_TOTAL_PLAYERS_0/_1 at 0x18020a0f8 / 0x18020a110
auctionCount(0x33) -> R+0x38 -> TEXT0 of TILE_ID 0x1b0, the TRANSFERS tile
FUN_180139610 (the hub body parser, 14855 chars, censused in full: 18 atoms, none
missed) is the ONLY writer of +0x3c anywhere in the image, one write, guarded by
`if (iVar6 != 0x90)`. This is not a candidate, it is the field.
I SPENT A DAY ON THE WRONG SURFACE AND WROTE THE WRONG CONCLUSION. REBUILD_RESEARCH
S19 declared the counter "not server-fixable" with a mechanism that was internally
correct and completely beside the point: the tile never read the club-stat store.
Two things reinforced the error and both are now fixed in the docs:
* ENDPOINT_MAP said this response "uses C++ reflection / vtable dispatch, NOT an
inline atom ladder -- no static field ladder to read" and marked it a GAP. False.
There is an inline ladder, one indirection away.
* The eight-row MY CLUB panel was assumed to be FUN_180043b90 case 1, which
publishes six keys, and I treated the six-versus-eight mismatch as a puzzle rather
than as evidence. It is a DIFFERENT provider, FUN_180094ce0, using a different
string family (FUT_MYCLUB_*), reading neither the mode tag nor any type id we were
sending. Two providers; we were reading the wrong one.
That is the third negative claim of this shape to fail today, after "this deserializer
has no skip handler" and "the factory does not wipe the stat map".
auctionCount is included as a FREE CONTROL: different field, different tile, so if MY
CLUB moves and TRANSFERS does not, delivery is fine and something is specific to +0x3c.
Default ON. Freeze risk is low by construction rather than by belief: a flat object of
two integers, both read with the INT getter, so there is no array, no nested object and
no type-desync surface. FUT_HUBDATA=0 restores {}.
Contract guard added, and verified to bite rather than merely pass:
default 439 checks, 0 failed
FUT_HUBDATA=0 435 checks, 3 FAILED (clubPlayers missing / not a number)
A regression here would otherwise be silent: still 200, still valid JSON, tile quietly
back to 0.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
The pool was 18 players rated 85 to 94. open_pack() split it with `(rating >= 75) ==
gold`, so the bronze pack's filter matched NOTHING and fell back to the whole pool:
all three packs dealt gold rares and the bronze pack was a lie. The file's own TODO
asked for "a full dbdata.dll extract (~18k players)".
NEW fut_cards.py: 79 players, 39 gold / 20 silver / 20 bronze, 7 leagues, 20 nations,
18 teams, every outfield position plus GK, no duplicate asset ids. PACK_CATALOG now
carries a weighted `tiers` draw per pack. Simulated 40 opens of each:
Bronze Pack {'bronze': 159, 'silver': 41} 39 distinct cards, 12 positions
Gold Pack {'silver': 110, 'gold': 170} 59 distinct cards, 12 positions
Premium Gold {'gold': 392, 'silver': 48} 57 distinct cards, 12 positions
WHY THIS IS NOT THE dbdata EXTRACT, and why that does not matter yet. dbdata.dll is a
real PE with one export, getTableData, whose 2.5MB payload sits in a section named
.xdata that disassembles as obfuscated code rather than a table directory, so the base
DB is not statically extractable without running that export under Wine or defeating
the obfuscation.
More to the point it would change nothing on screen today. Per docs/CARD_SYSTEM.md the
card view-model 0x1800d7920 reads EVERY rendered field (rating +0xb4, position +0x146,
nation +0x148, teamid +0x94, six attrs +0x98..0xac, name +0xdd) from a definition
record resolved at item+0x10 out of the client's own CardsDb map, and NEVER from our
item JSON. Offline that map is EMPTY, so every lookup misses and a blank record is
emitted. No assetId we send, real or invented, can produce a named card until that map
is populated. That is a separate job (CARD_SYSTEM options A/B/C) about the CLIENT's
map, not about our pool.
What the pool DOES control is everything the server is source of truth for: the
gold/silver/bronze split, leagueId/nation/teamid which are exactly what the club-stats
drill-downs read (those now work, S20, and were being fed 5 leagues from 13 nations),
preferredPosition which decides whether a squad can be filled at all, and the six
attributes behind the market filters.
Asset ids: 18 are genuine FIFA 17 ids and are listed in VERIFIED_ASSET_IDS. The rest
are structural, and the docstring says so plainly rather than passing them off as real
players. Because the CardsDb map is empty offline an id being wrong has no visible
effect today; if the identity work lands, that set is the diff target.
Backwards compatible: open_pack() still honours the legacy `gold` boolean when no
tiers are given, and the old list survives as _LEGACY_POOL for the starter squad.
392 + 61 checks green.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
MY CLUB -> ENGLAND -> Premier League now reads 17. First non-zero number ever rendered
on that screen. Nothing froze, no other screen changed, 392 + 61 checks green with the
flag defaulted on and no env override.
S19 concluded "the MY CLUB counter is not server-fixable". That was too broadly scoped.
The nation and league drill-downs ARE server-fixable and are now fixed. S20 records the
corrected scope.
What made it work, from FUN_180043b90 case 3:
uVar7 = (**(param_2 + 0x18))(param_2, row, "LEAGUE_ID") THE UI ROW'S OWN ID
bronze/silver/gold = (+0x7f8)(store, uVar7, 2 / 3 / 4)
publish("PLAYERS_EMPLOYED", gold + silver + bronze) COMPUTED, never read
rare/kits/badges = (+0x7f8)(store, uVar7, 5 / 0x28 / 0x2d)
Still open and now correctly scoped: the hub tile's "0 TOTAL PLAYERS" and the MY CLUB
summary rows read the GLOBAL bucket via +0x800 in cases 1 and 5. We serve those rows.
The unchanged question is what SELECTS those cases, since the mode tag is copied from
the completed request and the client requests year, consumables, staff, country/<id>
and league/<id> but never club.
The method note, which is the durable part: two rounds of reasoning about this endpoint
produced two wrong bodies; twenty lines of the consumer produced the right one. Reading
the PARSER tells you what is accepted. Only reading the CONSUMER tells you what is used.
That question was answerable from the start and went unasked until live screenshots
forced it.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
Second correction in an hour, and this one comes from reading the provider instead of
reasoning about it. The per-context getter is (+0x7f8)(store, contextValue, typeId),
and contextValue comes from THE UI ROW, not from the URL:
case 3: uVar7 = (**(param_2 + 0x18))(param_2, row, "LEAGUE_ID")
bronze = (+0x7f8)(store, uVar7, 2)
silver = (+0x7f8)(store, uVar7, 3)
gold = (+0x7f8)(store, uVar7, 4)
publish "PLAYERS_EMPLOYED", gold + silver + bronze
rare/kits/badges = (+0x7f8)(store, uVar7, 5 / 0x28 / 0x2d)
case 4: keyed by "TEAM_ID"; reads 1 (players), 0x28 (kits), 0x2e (badgeDBid)
Three things my previous commit got wrong:
1. It keyed every row to the id in the URL. The reader iterates the SCREEN'S ROWS and
looks up each row's own id, so one response must carry a bucket per row. Keying to
the URL id fills exactly one bucket the screen never asks for, which is why the
ENGLAND tab still showed zeros after the "fix".
2. PLAYERS_EMPLOYED is COMPUTED as gold + silver + bronze in the per-context cases and
is never read from the store, so sending `players` (type id 1) does nothing there.
The tier counts are mandatory.
3. The screens NEST: country/<id> lists the LEAGUES in that nation (case 3, LEAGUE_ID)
and league/<id> lists the TEAMS (case 4, TEAM_ID). That matches the live navigation
exactly: selecting ENGLAND produced Premier League / Championship / League One /
League Two.
Because every response wipes the whole map, each response only needs its own screen's
buckets, which also avoids a real collision: the storage key is contextValue alone, so
nation 14 and league 14 would otherwise share a bucket.
Live output now:
country/14 -> 41 rows, 5 league buckets
league 13 gold=17 -> PLAYERS_EMPLOYED=17 (Premier League)
league 19 gold=26, league 53 gold=55, ...
league/13 -> 6 team buckets {5:20, 21:15, 22:11, 240:21, 241:30, 243:17}
Recorded as a method note: two rounds of reasoning about this endpoint produced two
wrong bodies, and reading twenty lines of the provider produced the right one. The
question "what does the reader look up" is answerable and was not asked.
392 + 61 checks green, zero tracebacks. Still behind FUT_CLUBSTATS, default off.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
LIVE 2026-08-04, and this corrects the body I shipped an hour ago. Selecting the
ENGLAND tab on the MY CLUB screen issues exactly one request:
14:30:32 GET /ut/game/fifa17/club/stats/country/14 (14 = England)
and NO item-list request. So that tab is driven entirely by per-nation stats, and it
showed nothing while the club holds 8 England players.
THE GUARD WAS THE BUG. In deser 0x180130150, contextId == 1 or 5 <= contextId <= 9
FORCES contextValue to 0, which is the global bucket that the +0x800 getter reads. The
per-nation view reads the +0x7f8 getter keyed by the NATION ID instead. Every row I
sent carried contextId 1, so no matter what contextValue said, everything landed in
the global bucket and the per-context tabs could never see it. I had the guard written
down in my own comment and still sent a body that tripped it on every row.
Now, for country/<id>, league/<id> and team/<id>, the response carries the global rows
AND per-context rows keyed by that id, computed from the club's real nation/leagueId/
teamid fields:
country/14 -> players 8, playersGold 8, rarePlayers 8, silver/bronze/kits/badges 0
Both sets ride in the SAME response because every response wipes the whole map first,
so anything left out is erased rather than merged.
contextId 3 is used purely because it is OUTSIDE the guard and therefore preserves
contextValue. TODO/CONFIRM what contextId means semantically; nothing read so far
gives it a meaning beyond that guard.
Also verified rather than assumed this round: all 11 type strings we emit resolve
correctly against docs/fut_atoms.tsv (players 0x238, rarePlayers 0x272, stadia 0x2d7,
balls 0x4f, kits 0x17c, badges 0x4b, trophies 0x340 ...), 0 mismatches. So the strings
were never the failure.
Does NOT claim to fix the MY CLUB hub counter, which remains the open question in S19.
This fixes the nation/league tabs, which is a different and now-understood symptom.
392 checks green. Still behind FUT_CLUBSTATS, default off.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
Two experiments, both negative, and the negative has a mechanism behind it rather than
being another failed guess.
FUT_CLUB_PAGE served 114 items to /club?...count=11 tile still 0 TOTAL PLAYERS
FUT_CLUBSTATS full stat set, players=114, all modes panel still Players 0
The bodies went out: four "CLUBSTATS: 11 stat rows (players=114)" responses in the log,
panel re-entered afterwards.
FUN_18012fbe0 store+0x78 = request+0xc4 the mode tag is copied from the
REQUEST that just completed
FUN_180043b90 switch (store+0x78)
case 1 +0x800(1)->PLAYERS_EMPLOYED, (0x1e)->BALLS_EARNED, (0x28)->KITS_AVAILABLE,
(0x14)->STADIA_OWNED, (10)->STAFF_EMPLOYED, (0x32)->TROPHIES_WON
case 6 CARDS_NO_TRAINING_*, CARDS_NO_CONTRACT_*, CARDS_NO_FITNESS_*
The MY CLUB summary is case 1, which needs mode 1 (club). The client requests staff,
year and consumables and NEVER club, so the tag settles at 6 and case 1 is never
selected. Our values are stored correctly (contextId 1 forces contextValue 0, the
global bucket the +0x800 getter reads) and case 1 reads exactly the six ids we set.
Nothing ever asks for them.
THE MODE IS CHOSEN CLIENT-SIDE FROM THE REQUEST URL. No response body can change it,
so there is no body that fixes this and generating more of them is wasted work.
Two independent corroborations rather than one story that merely fits:
- case 6 reads 0x3d CONTRACTS, 0x3e TRAINING, 0x40 FITNESS, exactly the three ids
FUN_18012fd40 cannot produce from any type string. The consumables view is
unsettable from this endpoint by construction.
- FUT_CLUB_PAGE eliminated the only other candidate: the tile is not a count of the
list we return.
Both flags stay implemented and default OFF. FUT_CLUBSTATS is correct against the
verified schema and would populate the moment a club-mode request occurred; deleting it
would throw away the schema work for no gain.
Recorded against myself: I argued from the matching labels (tile "TOTAL PLAYERS", panel
"Players", both zero while we served {}) that the two read the same store and one body
would fix both. The store IS shared. The SELECTION is not, and that is what decides it.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
LIVE 2026-08-04: the draft-state array fix WORKED. The screen rendered instead of
hanging and the client advanced to the entry-fee screen, then crashed on the next
call, which we had never implemented:
GET /squad/mode/draft/state?mode=ONLINE -> our array body screen RENDERED
GET /user/credits -> 7200
GET /store/purchasegroup/all -> the entry-fee screen
POST /purchase/mode/0/draft {"currency":"COINS","usePreOrder":0}
-> {} UNMAPPED, then the crash
Advancing the failure to the next unimplemented call is what a correct fix looks like.
ENVELOPE AMBIGUITY RESOLVED, and ENDPOINT_MAP's note about it is wrong. Two structures
reference RS4:FutPurchaseDraftModeServerResponse:
0x18014c260 vtable 0x180224ef8, factory 0x18014c090. 3188 chars, OBJECT root
(prologue tests != 10 = END_OBJECT), 1 skip handler, exactly the seven
scalar ints. THIS IS THE RESPONSE PARSER.
0x180150310 vtable 0x1802262f0, factory 0x180150260. 1836 chars, ARRAY root
(loops until 0xd), ZERO skip handlers -- and NOT a response root at
all. It parses ENTRANCE CRITERIA: each element's name is strcmp'd
against the literals "COINS", "POINTS", "DRAFT_TOKEN" and stored at
+0x28/+0x2c/+0x30. It shares the class-name string because it is the
fee sub-object, not an "alternate/summary envelope" as documented.
THE CRASH ITSELF DISCRIMINATED, which is worth keeping as a technique. An object-root
parser handed {} parses benignly and leaves defaults; an array-root parser handed {}
desyncs and HANGS, which is exactly what draft/state did before the fix. We observed a
CRASH, not a hang, so the object-root parser is what ran and the failure is downstream
of an empty-but-valid parse. Consistent with 0x18014c260, inconsistent with the other.
Coins are NOT deducted. The client posts the price in the URL and it sent 0, because we
omit entranceCriteria from draft/state so there is no fee to charge. Charging a guessed
amount would be inventing an economy rule.
ALSO FIXED, before it reached the game: the route table hands handlers the compiled
PATTERN, not a match object (the dispatcher calls fn(rx, self)), so calling .group() on
the first argument raised AttributeError and killed the connection outright. That is
strictly worse than the {} it was replacing. Caught by verifying the response actually
changed after the restart rather than assuming the route worked.
Default ON: the behaviour it replaces is a confirmed crash, so no working state is at
risk. FUT_DRAFT_PURCHASE=0 reverts.
392 + 61 checks green, zero tracebacks on a clean boot.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
Live 2026-08-04: the CLUB STATS panel shows eight zeros (Rare Players, Players, Staff
Employed, Stadia Owned, Trophies Won, Kits, Badges, Balls Earned) while the client
fetches /club/stats/{staff,year,consumables} and we answer {} to all three. Those
zeros are ours. Every row name maps to a type string in the recovered map.
Wire schema, fully verified from deser 0x180130150 (7,870 chars, read end to end):
{"stat":[{contextId:int, contextValue:int, type:string, typeValue:int}]}
Unknown keys route to FUN_180135ff0 at BOTH levels, so extras are inert.
FIVE THINGS THAT DECIDE WHETHER IT WORKS:
1. EVERY RESPONSE WIPES THE WHOLE MAP FIRST. Nothing accumulates, so a good body on
one mode followed by a thin one on another ERASES the first and request ordering
decides what survives. Handled by serving the SAME COMPLETE SET on every Stats2
mode: whichever lands last leaves the map correct. (One investigator reported this
factory does not wipe; a reviewer re-read it and refuted that. The wipe is real,
and this is the second negative claim from that batch to fail.)
2. /club/stats/staff IS A DIFFERENT CLASS: FutStaffBonus, {"bonus":[{type,value}]},
not Stats2. Its type strings are undecoded so it keeps {}, which is safe and also
means it does not disturb the Stats2 map.
3. ELEMENT-LOCAL VARIABLES ARE NOT RESET BETWEEN ELEMENTS -- the clears sit before
the array loop, not inside it -- so omitting a key in element N inherits element
N-1's value. All four keys are emitted in every element.
4. The storage key is contextValue ALONE; contextId is only a guard (1, or 5..9,
forces contextValue to 0, the global bucket the +0x800 getter reads). contextId 1
throughout.
5. 0x3d CONTRACTS, 0x3e TRAINING and 0x40 FITNESS are READ by the panel but cannot
be SET from here. No type string produces them.
THIS IS ALSO NOW THE HUB-TILE CANDIDATE. The investigation concluded the MY CLUB tile
does not read this store, but flagged that negative as BOUNDED: the interface comes
through a QueryInterface adapter, so the vtable is assembled at runtime and cannot be
read statically. Live evidence points the other way. The tile reads "0 TOTAL PLAYERS"
and the panel reads "Players 0" -- same quantity, both zero, both while we answer {}.
And FUT_CLUB_PAGE ruled out the alternative: 114 items served to /club, tile still 0,
so it is not a count of the list. Strong inference, not proof; this flag is the test.
The test is unusually clean: the club holds 114 items and all of them are players, so
every other row is an honest zero. If it works, exactly two numbers move (Players and
Rare Players, 0 -> 114) and nothing else changes.
Gold/silver/bronze thresholds are FIFA's rating convention (75+/65-74/under), not
something read out of the binary, and the code says so.
Default OFF. 392 + 61 checks green.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
Five parallel Ghidra investigations, one adversarial reviewer each, one synthesis.
Kept in the repo because the reviewer corrections are load-bearing: they refuted claims
in four of the five reports, two of which would have shipped wrong behaviour (a
speculative /season body justified by our own curl traffic in the log, and a store
field block that was a freeze rather than a regression).
Carries the next live session (one launch, three flags, four menu actions, one
read-only memory probe), the implementation queue, what is genuinely blocked and why,
and a what-could-make-this-plan-wrong section that names the store change as the
concrete regression risk.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
This file has been handing out bodies that freeze the client, under the heading
"MINIMAL known-good".
1. FutGetDraftCurrentState. The root container is a JSON ARRAY. The documented body was
object-root, used the spelling DRAFTSQUAD_ON which is NOT an accepted squadState
value, and embedded a full squad object. Anyone serving it would have reproduced the
exact hang the entry existed to prevent, which is what happened live on 2026-08-03
when our generic /squad route answered this endpoint with a squad object.
Path corrected too: it is ut/%s/squad/mode + /draft/state, not ut/%s/draft/state.
The `squad/mode` segment was missing, which is why the URL is invisible to the
request-template table. Established by live capture, not statically: FUN_180146ac0
appends the suffix to a caller-supplied buffer and has no resolvable callers.
2. FutGetDraftAward (0x1801510c0) has the same array-root prologue, and its documented
object-root body would hang identically. Corrected, and marked TODO/CONFIRM on the
member list, which was not re-verified this pass.
Both of these survived because a census claimed only three array-root readers
existed in the DLL. It missed one. The census run to check it was wrong in the other
direction. ~23 of 86 top-level readers are still unclassified, so the file now says:
do not serve any endpoint here until its root container is classified by reading the
actual prologue, not by regex.
3. roundsInfo element: `score` and `penaltyScore` offsets were swapped (+0x10 / +0x18).
4. FutSeasonList: deserializer is 0x1801683f0, not 0x180167740 (that is the ELEMENT
parser), and the root is an OBJECT with one key `seasons`(0x2ad), not an array.
Someone documented the element parser's key set at the document level, and
utas_server.py served that shape for months on the strength of this row. Three of
the listed element keys are inner members of elgReq and inert at element level.
Added: element ordering (type before divisionId), stride 0x318, the (0xb-divisionId)
short, and the three array-loop members that must stay omitted.
5. Recorded on the season entry that the client has NEVER requested /season across 486
real requests, so no body there is observable yet.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
ZERO-LAUNCH FIXES from the multi-agent pass over 0x18013af30 and 0x1801683f0.
FUT_STORE_FIELDS IS DELETED, AND IT WAS A FREEZE, NOT A REGRESSION. It sent
"actionType": 0 and "firstPartyStoreId": 0 as JSON INTEGERS. Both atoms (0x8, 0x127)
are read with the STRING getter 0x1801c7aa0. That is the exact type-desync class this
project exists to avoid. So "the corrections stopped packs opening" was never bad luck
or an unrelated field: two of them were the documented freeze mechanism, shipped by a
change whose own comment claimed it was correct about what the parser reads. Knowing
WHICH atoms a parser reads tells you nothing about which TYPES it demands. Read the
getter, every time. (Two other fields in that block were no-ops anyway: useDefaultImage
0x36a stores inverted, and visible 0x37d never reads its value at all.)
FUT_STORE_GROUPS IS DELETED and its freeze is now traced end to end. It sent
displayGroup as an ARRAY of pack-shaped objects on the belief that the key was parsed
recursively by the same element parser. It is not recursive at all: case 0xd9 never
re-enters 0x18013af30. It is a FLAT OBJECT with exactly two members. An array desyncs
the reader, the parser runs off the end of the document, and the tokenizer returns the
same token forever with nothing consumed, spinning inside FUN_1801c7f10 whose body
contains the observed PC 0x1801c7f1a.
Both were being kept as togglable "maybe nearly right" experiments. Each is a loaded
gun; neither survives contact with the decompile. Deleted rather than left switched off.
NEW FUT_STORE_DISPLAYGROUP (default 0): send the one key that actually names a tile,
displayGroup = {"value": "<pack name>"}. `value` (0x377, STRING) writes record offset
+0x00, the same slot whose constructor default is the literal "unknown" (the only such
literal in the DLL, 0x180223108). The tiles say "unknown" because nobody ever sent the
field. Distinct values per pack so grouping stays 1:1. priority/displayGroupAssetId/
displayGroupUseDefaultImage all omitted as second variables.
Default OFF for a reason the token-balance proof does not cover: this may be the first
field we have sent that selects a RENDER PATH rather than a value, and that code is
packed.
SEASON ROOT SHAPE CORRECTED. season_list() and its docstring were both wrong in the
same way: the deserializer is 0x1801683f0 (object root, one key seasons=0x2ad, array
inside), not 0x180167740, which is the per-ELEMENT parser. Someone read the element
parser and served its key set at the document root, so a bare array populated nothing.
Three of the keys served (eligibilityKey/Slot/Value) are inner members of elgReq and
inert even at element level. Also recorded: `type` must precede `divisionId` because
the divisionId branch reads the parsed type at elem+0x1b4.
Still behind FUT_MODES and still pointless to serve: across 486 real client requests
the game has NEVER asked for /season. Every /season line in our log is our own curl.
NEW FUT_CLUB_PAGE (default 0): an experiment, not a fix. The MY CLUB counter's
renderer is not in cardsdll (no two-number formatter of any spelling exists) and
FutStickerBookSearch has no count atom, so there is no field we can send that IS the
number. What is still testable is whether the counter is a Flash-side count over the
returned list, which a pure length change discriminates. A NULL RESULT IS THE VALUABLE
ONE: if the counter does not move, there is no server-side lever for this symptom and
the right outcome is to prove that and stop.
392 + 61 checks green, defaults byte-identical.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
The "4-byte header" that precedes every response class name in .rdata, which cost six
failed class-to-deserializer resolutions before anyone noticed the offset, is not a
length prefix or a refcount. It is the string RS4:. The full literal is
RS4:FutXServerResponse, and searching for the bare class name lands four bytes in.
Verified directly on three classes:
FutDestroyMatchServerResponse name@0x18021d694 header = b'RS4:'
FutGetDraftCurrentStateServerResponse name@0x180224204 header = b'RS4:'
FutStickerBookStats2ServerResponse name@0x1802220cc header = b'RS4:'
Found by a verification agent that had been instructed to distrust the rule. It did,
and came back with the reason rather than the offset. A magic constant you have to
remember is a rule you will eventually get wrong; a prefix you can read is not.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
The whole match family is one RPC descriptor block (rows 49-54, every row using URL
template index 16 = `ut/%s/match`) with a fixed suffix appended per call:
CREATEMATCH ut/game/fifa17/match PLAYGAME ut/game/fifa17/match
MATCHREADY ut/game/fifa17/match/ready DESTROYMATCH ut/game/fifa17/match/end
RESETMATCH ut/game/fifa17/match/reset KEEPALIVE ut/game/fifa17/match/keepalive
THERE IS NO /match/{id} URL. The id travels in the body. Our reward path was gated on
`h.command == "DELETE" or "/ut/delete/" in h.path` and extracted the id with
re.search(r"/match/(\d+)"), so it was waiting for a request the client does not make.
The gate is now widened to include a /match/end path with ANY verb, because the verb
genuinely cannot be determined statically: the strings "PUT" and "DELETE" do not exist
anywhere in cardsdll.dll (0 hits each), so verb selection happens outside this DLL. A
reviewer flagged "the reward path can never fire" as overreach on exactly that point;
widening rather than replacing the gate is the response.
THE RESULT SIGNAL IS `endReason` (atom 260), a STRING enum with nine values: WIN DRAW
LOSS DNF QUIT NO_CONTEST DNF_WIN DNF_DRAW DNF_LOSS. Not a score comparison. The score
lives in `myMatchStats.goals` / `opponentMatchStats.goals`, two literal-keyed objects
of 15 int fields each, and the client OMITS both when endReason is DNF or QUIT, so
nothing may require them. _match_result() now reads endReason first and keeps the old
spelling probe only as a fallback, because request-side static findings are a floor:
PUT /item's swap/tradeId appeared in no static listing either.
THREE CORRECTIONS TO THE RESPONSE, all of which were shipping wrong:
1. `coins` (atom 149) is NOT a top-level key. It is read only inside `gameModeAward`.
The one field most obviously named "the reward" was being silently skipped.
2. `qualifiedChampionEventId` (0x269) has a SIDE EFFECT: its branch calls through a
manager vtable after storing. Sending a habitual zero poked champion-event
machinery for no benefit. Removed.
3. `bidTokens` (atom 89) inside gameModeAward is MATCHED and then handled by nothing,
so its value token is left unconsumed. That is the precondition for the desync
spin. A freeze trap dressed as an ordinary field; now guarded by a unit check.
test_match_rewards.py rewrote its expectations. The old version asserted a top-level
`coins` and passed happily while the server shipped a body whose reward field the
client never read. A test that encodes the wrong schema converts a bug into a
guarantee. New test_end_reason_is_authoritative covers all nine enum values, the
stats-less DNF case, and that endReason beats a contradictory score probe.
DEFAULT ON (FUT_MATCH_END=0 reverts), a reasoned exception to the flag convention:
nothing here is live-proven because no match has ever been played, and the old
behaviour is not a working screen but a path that provably could not fire.
61 unit checks (58 with the flag off), 392 contract checks green.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
Deser 0x180147070 (FutGetDraftCurrentStateServerResponse) discards tokens until it
sees START_ARRAY. Handed a top-level OBJECT it never reaches its exit condition and
spins in the inner `while (tok != END_OBJECT)` loop while the tokenizer returns EOF
forever. Process alive, no crash dump, no dialog: exactly the signature observed live
on 2026-08-03, when our generic /squad route answered this endpoint with a full
active-squad object.
The suffix composition `ut/%s/squad/mode` + `/draft/state?mode=...` makes the URL
invisible to the request-template table, which is why /squad swallowed it. Fourth time
a suffix endpoint has been invisible to that table, second time the generic /squad
route has eaten one (/squad/list was the first).
Verified at instruction level rather than by regex: 7 top-level atoms (4 int-getter
calls, 3 string-getter, 0 bool, 2 skip) across the whole body [0x180147070,
0x1801475a3]. FUN_180135ff0 IS present, twice, so unknown keys are inert. NO ATOM
COLLIDES with the squad object we were serving, which means the hang was purely the
container level and not a per-field type desync.
entranceCriteria(0x108) is now known to be an object of three int keys
COINS/DRAFT_TOKEN/POINTS. It is OMITTED anyway: knowing a shape is not a reason to
send it.
A second agent independently simulated this exact body through the deserializer line
by line and got a clean exit in 16 token reads, and separately refuted four claims in
the first agent's report (a census undercount, a wrong .rdata address where
0x18021e7f4 is 'TFA' not the squad template, an incorrect stateParam2 typing argument,
and a dangerous aside about a second array-root envelope). The body survived all of it.
DEFAULT ON, a deliberate exception to "default to the live-proven value": the
live-proven value here HANGS THE GAME, and there is no working screen to protect
because Draft cannot be entered at all today. FUT_DRAFT_STATE=0 restores the old
routing.
392 + 51 checks green; /squad/0 and /squad/list verified unchanged.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
User-Agent filter: implemented as the default in futlog.py.
The two wrong .rdata addresses: verified they never reached any document. They existed
only in an agent recon report, so there was nothing to correct. Kept the reviewer's
corrected values because they are verified and useful:
RS4:FutGetClubInfoServerResponse 0x180221a38 (not 0x180220e38)
RS4:FutStickerBookSearchServerResponse 0x180221e48 (not 0x180221248)
Closing an item by checking it was never a problem counts as closing it.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
/leaderboards/options is the ONLY mode-related endpoint the real client has ever
requested, and we answer {} because FUT_MODES is off. Three of its seven occurrences
are followed within ~2 minutes by /user/accountinfo and a fresh /ut/auth, which is the
signature of hitting an error, returning to the main menu, and re-entering FUT.
Hypothesis: the client fetches mode options on entering the play area, caches them, and
later refuses Seasons from that cached EMPTY body without issuing another request. That
would explain the zero-requests-at-failure observation, which no response-shape theory
has been able to account for: the deciding fetch happened minutes earlier.
Stated as a hypothesis, not a finding. The correlation is real; the causation is not
established. Cheap to test: leaderboard_route already implements an options body behind
FUT_MODES=1.
Risk noted in advance: FUT_MODES=1 also enables /season, whose array-root shape is a
flagged freeze candidate. That risk cannot fire while the client never asks. If this
hypothesis is right, a populated options body is precisely what would make it ask for
the first time, so succeeding at step one arms step two.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
The bug we just closed was invisible to every existing check: PUT /item answered 200
with a well-formed JSON body, and the body told the client the move had FAILED. Seven
attempts, weeks of investigation, and nothing in the suite would have noticed a
regression back to {}.
test_move_verdict_shape asserts the record vector exists, has ONE RECORD PER REQUESTED
ITEM (an empty vector fails the client exactly as hard as a wrong field), and that
id/pile/success carry the number/string/bool types their getters require.
Non-mutating: it asks to move ids that cannot exist, so nothing changes pile. The
verdicts come back success=false, which is honest and is not what is asserted.
Verified it actually catches the regression rather than just passing:
default (ack) 392 checks passed, 0 failed
FUT_MOVE_BODY=empty 382 passed, 1 FAILED -- "returns itemData array"
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
REBUILD_RESEARCH S18, from running futlog.py over the full 3044-request history.
The client has requested /club six times and EVERY ONE carried a query string
(year/type/count/position/level/nation/league/team/sort). The bare path has never
been requested. ENDPOINT_MAP says GET ut/%s/club is FutGetClubInfo, whose only
recognised member is user(0x36c), and concludes our itemData body is skipped and the
club list must therefore be empty. The club list is NOT empty; every card renders. So
either the query form dispatches elsewhere or the row is wrong. TODO/CONFIRM.
Relevant to the MY CLUB counter: we ignore the query string completely and return all
109 items to a request asking for count=11 with position and sort filters, and our
response carries no result total. A paged search response is exactly where a tab
counter would read its number from.
Also recorded: the unmapped view is a standing detector for suffix endpoints the URL
template table cannot show. It has caught four so far.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
Implements the standing requirement recorded in priority-2026-08 S5: the User-Agent
filter is now the DEFAULT in the log tooling, not an option. The real client sends
ProtoHttp; our own probes send curl/* or Python-urllib/*. Reading the log unfiltered
gave this project a materially wrong picture of itself (/clubUser and /user/list had 93
and 180 hits, none from the game).
The old futlog.py was a one-off with a hardcoded path and no notion of who made the
request. Replaced with a real tool: timeline or summary, body/response display, path
regex, time window, status filter, and an --unmapped view that lists the endpoints the
client wants and we catch-all. --all and --probes exist for when you deliberately want
our own traffic.
Over the full 3044-request history: 486 requests came from the game.
IT IMMEDIATELY CAUGHT ME OVERSTATING SOMETHING. Yesterday's commit called the PUT /item
request shape "captured for the first time (the client had never successfully reached
this path)". False. There are NINE client PUT /item requests in the log, eight of them
during the failed attempts, every one carrying swap and tradeId:
08:45:11 08:51:13 08:55:21 08:58:19 09:10:52 09:15:26 09:22:31 09:37:12 | 11:15:48
The request was on the wire and in the log the whole time. What was new was reading it.
Same class of error as the truncated decompile in S16: evidence already collected and
not looked at. REBUILD_RESEARCH S17 corrected.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
Live 2026-08-04 with FUT_MOVE_BODY=ack and FUT_PACK_AUTOCLUB=0. Bought a bronze pack,
opened it, chose Send to Club. The session SURVIVED, the five cards persisted into the
club pile, and there was no ut/delete/auth logout -- the logout that accompanied all
seven previous attempts.
11:15:48 PUT /item
req {"itemData":[{"id":100000125,"pile":"club","swap":0,"tradeId":0}, ... x5]}
res {"itemData":[{"id":100000125,"pile":"club","success":true}, ... x5]}
11:15:49 GET /user/credits session alive
11:15:51 GET /hub no error dialog
11:16:06 GET /club?year=2017... MY CLUB opened
PUT ut/%s/item never was an ack endpoint. It builds per-item VERDICT records, and the
completion handler raises EVENT_CARDS_MOVE_CARD_FAILURE when the vector is empty or
success != 1. Every body this project ever returned, {} included, told the client the
move had FAILED, and the client ended the FUT session because that is what that event
does. We were failing our own move.
Defaults flipped: FUT_MOVE_BODY empty -> ack, FUT_PACK_AUTOCLUB 1 -> 0. The autoclub
workaround is retired.
New intel, captured for the first time because the client had never got this far: the
request carries `swap` and `tradeId` beside id/pile. We ignore both and the move
succeeded, so neither is load-bearing for a pending-to-club move.
PROCESS, and this is the part worth keeping. The FIRST attempt at this test produced
no PUT /item at all: FUT_PACK_AUTOCLUB=1 had already emptied the pending pile at
purchase time, so the reveal screen had nothing to assign and the client never issued
the request. The workaround for the bug was hiding the bug. Before testing a fix,
check the configuration still lets the client make the call the fix is for.
Two self-inflicted incidents, both recorded in REBUILD_RESEARCH S17:
- Restarting the server to inject a flag WHILE FIFA was running produced the exact
"error connecting to FIFA 17 Ultimate Team" dialog this project spent weeks chasing,
from a plain connection refusal during the ~30s window. Restart only at the main
menu, and check the log for ProtoHttp requests before blaming a response.
- pgrep -f matched the invoking shell twice, killing it before the restart, because
the same command contained the literal script name in a later clause.
Docs updated: REBUILD_RESEARCH S17 (the solve), priority-2026-08 S2/S3.1/S6 (next task
is now the MY CLUB counter), PROJECT_REPORT 6a, HANDOFF 5a plus the stale "FutMoveCard
has no skip handler" claim in S3 and a new 5d for Seasons/Draft.
380 + 51 checks green.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
The deliverable owed from the assessment brief. Ordered by cost of the measurement
that would settle each item, not by how important the outcome feels.
Headline reordering: the observation session ran and did NOT produce the match
shape. Seasons refuses with zero requests to any layer and Draft hangs, so both
routes into a match are blocked and /match is now behind a fix rather than in front
of one. The next task is instead one launch with FUT_MOVE_BODY=ack, which is the
only open problem where the decompiler has produced a verified necessary condition
that has never been satisfied.
Also recorded:
- The User-Agent split. Real client (ProtoHttp) vs our own probes. /clubUser and
/user/list have 93 and 180 recorded hits and not one came from the game. Every
"the client asks for X" claim predating the split is unsupported until rechecked.
- The narrowed-core measurement. The proposed boundary ("ownership and economy keyed
by opaque integer item ids") is correct and every per-item prediction held, but
narrowing is not what guts core: FIFA 17 relevance is. Six services totalling 869
lines model features with no FIFA 17 endpoint at all, survive narrowing perfectly,
and are worth nothing here. Suite: 61 of 101, not the 81 previously claimed.
Recommendation is to reuse the scaffolding and the 61 tests, not the service layer.
- Port timing stays "after", led by the match-result-shape argument: three of the
four queued items will change a response schema, and porting a placeholder schema
means porting the correction too.
- test_fut_contract.py is now implementation-independent (380 checks over HTTP), so
it can certify a Rust port. That is the port's main de-risking asset and it exists.
- A "What could make this plan wrong" section, including that a negative ack result
is not a refutation, and that every negative claim in ENDPOINT_MAP.md is weaker
than the corresponding positive one after the FutMoveCard retraction.
Standing requirements recorded, including the User-Agent filter default (required,
not yet implemented) and two wrong .rdata addresses still to correct.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
RETRACTION. This repo claimed, in REBUILD_RESEARCH S14c and in utas_server's
item_route comment, that FutMoveCard 0x180128600 "HAS NO SKIP HANDLER
(FUN_180135ff0 appears zero times, unique among FUT deserializers)" and "parses
only itemData -> dreamSquads". Every part of that is false. Full decompile:
FUN_180135ff0 call sites : 2 (offsets 5006, 6080)
atoms parsed : 7 active dreamSquads id itemData pile reason success
Cause: the decompile was written out as src[:4000] and then searched. The function
is 6193 chars, so BOTH skip-handler call sites and four of the seven atoms lay past
the cut. An absence was reported from a truncated listing -- the same failure mode
as the Memory.getBytes bytearray scan that silently returned zero hits. Never
conclude an absence without asserting the searched region covers the function.
Cost: the false claim implied "any extra key desyncs this parser", which sent the
investigation after client-side state for seven attempts, and the derived premise
"the deciding factor is client-side state, not the wire" was wrong too.
VERIFIED SHAPE. PUT ut/%s/item is not an ack endpoint; it returns per-item VERDICT
records:
id(0x15c) INT 0x1801c79d0 -> record+0x00
pile(0x226) STRING 0x1801c7aa0 -> enum 0x180142650 (club=7 purchased=6 trade=5)
success(0x2fa) BOOL 0x1801c7620 -> record+0x0c
reason(0x279) STRING -> "Destination Full" = 0xf
dreamSquads(0xe9) INT array; else -> FUN_180135ff0 (skip)
The completion handler raises EVENT_CARDS_MOVE_CARD_FAILURE when the record vector
is EMPTY or record+0x0c != 1, and success is initialised to '\0' per element. So
every body ever returned reported the move as FAILED, {} included. Quick sell
survives an identical {} because its callbacks read only the transport code and
ignore the body -- that is the whole asymmetry, and it was on the wire after all.
STAGED, NOT DEFAULTED. FUT_MOVE_BODY=ack emits the correct shape; the default stays
`empty` because sufficiency is untested. One launch settles it.
The ack is answered BEFORE the `if moved:` gate: under FUT_PACK_AUTOCLUB=1 the
cards are already in the club when the reveal asks to move them, so move_items()
returns nothing and a moved-derived body would be zero-record -- failing in exactly
the configuration ack exists to fix. Caught in review before it ran. success is
asserted only for ids that were moved now or are already in the club; anything else
gets an honest success:false rather than an invented verdict.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
test_fut_contract.py no longer imports ACCOUNT from fut_account. The expected
persona comes from FUT_TEST_PERSONA_ID and the target from FUT_TEST_BASE, so the
suite now imports nothing but stdlib and talks to a server at a URL.
That is what lets these 380 checks certify ANY implementation of the reversed
spec, a future Rust openfut-core included, without replaying the reverse
engineering. The original reason for reading ACCOUNT still holds and is preserved
in the comment: suite and server must not each hold a private copy of the
constant, or the identity-consistency checks would only prove two copies matched.
Also adds pileSizeClientData(0x227) behind FUT_PILESIZES (default off). A probe
run with 16 uniquely-valued entries did NOT move the MY CLUB counter, so that
member is eliminated as its source; the code is kept for the record and flagged
off.
Docs: OPENFUT_PROJECT_REPORT.md and OPENFUT_HANDOFF.md. The report now separates
"built but untested" from "never requested by the client" -- the server log records
User-Agent, and splitting real client traffic (ProtoHttp) from this project's own
probes shows /season, /tournament, /champion, /match, /clubUser and /user/list are
at ZERO client requests. /clubUser (0 client, 93 probe) and /user/list (0 client,
180 probe) are the starkest: work was done on both assuming the client wanted them.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
Second session. FUT core loop, the EASFC/POW online layer, a central account
backend, and a lot of corrections. Everything risky is behind an env flag with
the default set to whatever was live-proven.
WORKING END TO END (live-verified this session):
* match loop -- POST/PUT/POST/DELETE ut/%s/match, rewards via FutDestroyMatch
(0x180121b60). Play a match, get coins, W/D/L updates.
* packs -- buy, cards land in the club, session survives (FUT_PACK_AUTOCLUB=1)
* quick sell -- POST ut/delete/%s/item was UNMAPPED and paid NOTHING; six cards
were destroyed for 0 coins. Now credits discardValue.
* POW/EASFC online -- the "EA FC servers unreachable" banner is powdll's layer,
a THIRD http api on :8094 nobody had served. Redirect needs no root: powdll
FUN_18005a460 reads FIFA_POW_URL from the same client-config store as
ROSTERUPDATE_URL. FUT_POW=1.
* account backend -- fut_account.py replaces 7 hardcoded copies of the persona
across 5 files; club/persona/online-profile editable via CLI.
CORRECTIONS TO ENDPOINT_MAP (all re-extracted from the deserializers):
* FutStoreGetPackTypes: id/packType/isPremium/quantity/saleType/purchaseLimit/
purchaseCount are NOT skipped no-ops -- all are parsed. extPrice inner objects
take externalPriceId(0x11a), not amount/currency.
* FutMoveCard 0x180128600 has NO skip handler (FUN_180135ff0 appears zero times,
unique among FUT deserializers) and parses only itemData -> dreamSquads.
* class -> deserializer resolution: the name literal is preceded by a 4-BYTE
HEADER and the factory LEA points at the header, so look up name_addr - 4.
Six attempts failed on this; now ghidra_env.class_deser(). Unlocked 11 SBC/
Draft schemas.
* live-only endpoints the request table never lists: ut/%s/squad/list,
ut/%s/user/club, ut/%s/club/stats/*, ut/%s/clientdata/<key>. The template
table is a floor, not a ceiling -- the log is the only ground truth.
* 163 RS4 call names exist; we served 17. All now served.
FIXED: club/stats/* was answering with the entire 28-item club inventory on every
poll (it fell through to the generic /club route).
UNSOLVED: the pack reveal's "Send to Club" (PUT ut/%s/item) kills the FUT session
whatever we answer -- {} included -- while its sibling quick-sell endpoint accepts
a bare {}. Seven hypotheses eliminated by live test, documented in
REBUILD_RESEARCH.md S14c so none get re-walked. FUT_PACK_AUTOCLUB routes around it.
Also unfixed: store tiles render "unknown" (displayGroup is parsed RECURSIVELY by
the same element parser; sending it FROZE the store, so FUT_STORE_GROUPS=1 is
default off).
Tests: test_fut_contract.py 380 (live, read-only) + test_match_rewards.py 51 (pure).
Note: fut_store.py carries some pre-existing uncommitted changes from before this
session (pack catalogue ids, pending-pile behaviour) that could not be separated
from this session's additions in the same file.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
The client now issues PUT /squad and the hub renders coins, record and the
squad roster. Three separate root causes, all verified live.
Squad blocker (the long-standing "client never sends PUT /squad"):
AddPlayerToSquad, GetSquads and SelectSquadById issue ZERO network requests
(pure local model reads/mutations, FutSquadServiceImpl vtable 0x180233ff0);
only SaveCurrentSquad writes, and it is unguarded. The client simply needed a
populated ACTIVE squad model, which arrives via the massinfo `squad` member.
No response of ours was ever being rejected.
userMassInfo is NOT required to be {}:
0x180174630 is a FLAT {userInfo, squad, settings, userData} body -- the old
"wrapper key is user" note was wrong, and the historical freeze was the
malformed squad member, not the envelope.
clubNameChangeAllowed must be false:
sending true advertises a club-rename flow whose UI model is never populated;
the client shows a naming prompt and dies confirming it (ACCESS_VIOLATION
reading 0x0 at FIFA17.exe+0x71b8651, 4/4 runs, no CardsDLL frame and no request
in flight). Isolated by a single-variable run; guarded by a contract check.
Endpoint/schema corrections found in live traffic, invisible to static analysis:
* GET ut/%s/squad/list is a real endpoint and must return {"squad":[...]},
not the active-squad object (the /list suffix is appended by the caller, so
it never appeared in the request table)
* PUT lands on ut/%s/squad/<id>, not a bare ut/%s/squad
* userInfo currencies are read as name/funds/finalFunds/active -- there is no
"value" key, so coins always rendered 0
* squad-list elements take STRING formation/squadType, not ints
* the CardsDLL script-API thunk<->name table was off by one (AddPlayerToSquad
is 0x18004aa70; 0x18004aff0 is GetPotentialChemistry_Club)
FUT_MASSINFO / FUT_USERINFO ladders keep every step of the bisect reproducible.
Contract suite 311 -> 358 checks. Tooling added: PyGhidra harness (Ghidra's
Java/OSGi script path is broken on this box), minidump reader, live code grabber.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VUT92pz6RWKih9dSr8ZpxW
The FUT store 'not available' is FIFA's online-mode readiness gate
(FUT::CompetitionManager), not a config flag. tools/store_enable_poke.py finds
CardsDLL's live base and patches the 3 gate methods (0x1800f7fb0
IS_EASTORE_SERVICE_READY, 0x1800fb850 IS_STORE_ENABLED, 0x180100500
IS_COIN_PURCHASABLE) to 'mov eax,1; ret'. Reversible (saves originals; 'restore'
subcommand; FIFA restart clears it). Needs ptrace_scope=0 + FIFA running.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PN5bmpDVQR1aXgefyWAt7o
Live log ground-truth: FIFA's transfer-market SEARCH hits
GET /ut/game/fifa17/transfermarket?type=player&start=0&num=12 (was UNMAPPED ->
catch-all {} => empty market), NOT /auctionhouse as the struct name suggested.
Route it to the same listings handler. Market now serves 18 listings on the
endpoint FIFA actually calls.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PN5bmpDVQR1aXgefyWAt7o
Add enableSquadBuildingSetsFeature + FUT/SBC_USE_STUBS to the Blaze FUT config,
gated on env FUT_SBC (default OFF -> baseline unchanged, no restart needed). The
SBC set-list deser (0x180154990) checks FUT/SBC_USE_STUBS -- FIFA may render
built-in stub SBCs with zero server content. A concrete, safe lever to try SBCs
in the next live session without shipping complex (freeze-risky) SBC JSON blind.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PN5bmpDVQR1aXgefyWAt7o
Complete the market loop (browse + buy + sell). POST auctionhouse (FutISStart)
lists an owned club item -> profile.listings + returns {id:tradeId}. tradePile
builds a validated auction record per listing from the owned item + prices
(freeze-safe, same 0x18013e410 shape). DELETE trade/{id} removes the listing.
fut_store gains list_for_sale/listings/remove_listing (tradeId space 900500000+).
test_market_buy.py extended with sell/delist checks (temp profile, no real-save
mutation): list -> tradePile shows it with prices -> delist empties it. All pass.
Read-only contract suite still 311/311. Functional (FIFA's exact sell params)
pending live test; freeze-safe by construction.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PN5bmpDVQR1aXgefyWAt7o
trade_route now resolves the auction from its tradeId and, on buy-now
(bid >= buyNowPrice), spends coins + grants the won card to the club + echoes
the CLOSED auction (FutISOfferTrade shape). Reuses the validated auction record
(0x18013e410) so it stays freeze-safe; whether FIFA surfaces the won item
post-buy is functional (needs live test). Insufficient funds -> 461.
tools/test_market_buy.py: offline unit test on a TEMP profile (never touches the
real save) -- verifies coin deduction, card grant, closed-auction shape, and the
461 path. PASS. Read-only contract suite still 311/311.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PN5bmpDVQR1aXgefyWAt7o
Serve 18 real-player auctions on GET auctionhouse (search) built to the reversed
auction record schema (deser 0x18013e410) field-for-field: itemData reuses
fut_store._item (the proven club/squad card parser 0x18013fe00), scalar fields
all HIGH-confidence reversed. Rating-based buy-now pricing; tradeId space
900000000+. tradePile/watchList stay empty (no live sell/watch flow yet). Toggle
off with FUT_MARKET=empty.
Extend test_fut_contract.py to validate EVERY populated record field type
(numbers/strings/bool/object) so the listings are proven freeze-safe OFFLINE
before the game parses them. 311 checks, all passing.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PN5bmpDVQR1aXgefyWAt7o
Add tools/test_fut_contract.py -- stdlib-only, read-only tests that hit the live
utas_server and assert each response matches the shape reversed from CardsDLL
(docs/ENDPOINT_MAP.md). Encodes freeze-safety invariants (auctionInfo/players/
itemData/currencies/purchase must be array/object per the SAX deserializers;
scalar-where-container = busy-loop freeze at 0x1801c7f1a) plus the specific
contracts: v2/store gate == SUCCESS, coin counter binds currencies[coins].funds,
userMassInfo stays {}, empty squad slots carry itemData=null (proven-safe).
Catches the regression class that previously bit us (phantom packs, coins-0,
squad reset). 58 checks, all passing.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PN5bmpDVQR1aXgefyWAt7o
Add auctionhouse/trade/tradePile/watchList/marketdata routes per ENDPOINT_MAP
market §. All share the reversed IS-list body {auctionInfo:[], credits, total,
duplicateItemIdList} (shared deser 0x18013e7f0). Served EMPTY (no live listings
yet) -- empty arrays never desync the SAX reader, so freeze-safe; unlocks the
market screens vs the prior catch-all {}. GET auctionhouse merges the
FutGetAuctionCount ints (extra keys skip). POST auctionhouse = FutISStart new
tradeId; PUT relist / watch add-remove / delete = acks. tradePile precedes
/trade (prefix collision). Populating real auctions deferred to in-game test.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PN5bmpDVQR1aXgefyWAt7o
Store "not available" root cause reversed from CardsDLL:
- ut/v2/game/fifa17/store is an ELIGIBILITY gate (FutStorePackQuantities
deser 0x1801758c0), not a quantity list. It reads one key "result"
(atom 0x288); the store screen refuses to open unless SUCCESS. Was
unhandled -> catch-all {} -> "not available". Now returns {"result":"SUCCESS"}.
- Store-screen entitlement checks (0x18001749d/0x1800175a2) read IS_*/
*_PURCHASE_ENABLED Blaze flags, separate from storeEnabled. Added the full
confirmed set (14 flags) to FUT_RS4_CONFIG.
- Catalog: assetId (0x23) is the real pack identity; extPrice inner keys are
amount/currency (not mtx). (Also gated client-side by GetSystemMetrics>1024x768.)
Full FUT API reversed (clean-room, CardsDLL only) into docs/ENDPOINT_MAP.md:
~100 FutXServerResponse types across 7 feature groups (market, SBC, draft,
seasons/match, club, store, user/hub), each with deserializer VA, atom-mapped
field schema + types, freeze-risk flags, and minimal known-good JSON.
Tooling kept: tools/atomdump.py (dumps the 907-atom key table at 0x1802d2760)
-> docs/fut_atoms.tsv. Research prompt: docs/OPENCODE_ENDPOINT_PROMPT.md.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PN5bmpDVQR1aXgefyWAt7o
Reverse-engineered the exact FIFA17 store/purchase/credits response
shapes (wf a245577b + wf_76fcf89b) and applied them:
- Store catalog: root key MUST be "purchase" (atom 608) not
"purchaseGroups"; packs keyed by "id" (int16) not packId; price is a
"currencies":[{name,funds,finalFunds}] array; name is "description".
Our old {purchaseGroups:...} hashed to unknown atoms -> empty -> "store
not available". Now the store displays.
- Pack buy: gate open_pack on a transaction with "packId" and state !=
TRANSACTIONCANCEL (the TRANSACTIONCREATED create step) -- fixes the
phantom-buy. Reveal response = {"createPackResponse":{itemList,
numberItems,purchasedPackId,duplicateItemIdList}}
(FutCreatePackServerResponse).
- Coins: /user/credits must return currencies[name=="coins"].funds, not
{"credits":N} (the hub/store read currencies). squad_route also
reconstructs the active squad from club item-id references.
Verified via curl: store shows 3 packs; cancel spends nothing; Bronze
buy awards 5 real players and deducts 400 coins. NOTE: the FUT HUB coin
counter reads from userMassInfo (not /user/credits) -- still blocked on
the userMassInfo-freeze wall (separate reverse in progress).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PN5bmpDVQR1aXgefyWAt7o
FIFA's updateActiveSquad PUT stores each slot as itemData={id:<clubItemId>}
(a reference), not the full player. We saved the bare references, so the
squad reloaded empty ("active squad resets"). Add Store.reconstruct_squad()
to re-embed the full club item by id on GET /squad/0, so the saved squad
reloads with its real players.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PN5bmpDVQR1aXgefyWAt7o
store/transaction receives {"state":"TRANSACTIONCANCEL"} (cancel/close)
and {"packId":N} (pack-details fetch on store load) -- neither is a
confirmed purchase. The handler treated packId as a buy and even
defaulted to opening a Gold Pack on cancels, silently spending coins.
Make store_buy a safe no-op that logs every body, so the real
purchase-CONFIRM signal can be identified from a deliberate in-game buy
and open_pack() gated on exactly that. Save restored on next fresh run.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PN5bmpDVQR1aXgefyWAt7o
Add a first-cut FUT store on top of the persistent profile:
- fut_store: PACK_CATALOG (Bronze/Gold/Premium), a curated real-player
PACK_POOL, and open_pack() (deduct coins -> generate items -> add to
club -> persist).
- utas_server routes: GET store/purchasegroup/all (catalog), PUT
(v2) store/transaction (buy + open, returns awarded itemData +
updated coins), GET purchased (last pack). Matches both /ut/game and
/ut/v2/game prefixes.
Verified via curl: buy Gold Pack -> 7 real players awarded, coins
15000->10000, club 10->17, persisted. Wire format is a best-guess
grounded in the CardsDLL store keys (packId/price/itemData/coins);
iterate against the in-game store next. Pool is curated for now --
replace with a full dbdata.dll extract later.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PN5bmpDVQR1aXgefyWAt7o
Add fut_store.py: a JSON-backed profile store (coins, owned club items,
saved squads, record). First run grants a starter pack -- 15000 coins +
a 10-player starter club (real assetIds; identity resolves locally
in-game per docs/CARD_SYSTEM.md).
Wire utas_server to the store: /user/credits -> persisted coins,
/club -> persisted owned items, PUT /squad -> persists the squad the
user builds so it survives relaunches. Profile save file is gitignored.
Foundation for pack-opening (store/purchasegroup + transaction) next.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PN5bmpDVQR1aXgefyWAt7o
A full 88-rated real FUT squad (Ronaldo/Messi/Suárez/Ramos/Kroos/Alba/
Hazard/Oblak/Alaba/Boateng) renders 100% offline, no EA servers.
Mechanism: the definition fetch was unnecessary — FIFA has all player
identity locally in dbdata.dll. The CLUB-SEARCH / Add-Player flow
(GET /club?type=player&count=N, served by our /club route) makes FIFA
resolve each item's assetId against its own local DB (real name/photo/
club/nation) and merge the rating/attributes our /club item carries,
caching a real record in the CardsDb store. No idList fetch, no leaked
data.
Recipe: utas FUT_SQUAD_STEP=s3v0 -> /club serves the full XI; in FUT,
Squads -> Add Player/search -> results render real -> add to slots.
docs/CARD_SYSTEM.md updated with the SOLVED mechanism + recipe.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PN5bmpDVQR1aXgefyWAt7o
Milestone: FIFA 17 Ultimate Team boots end-to-end on our offline backend
past every EA gate into the hub and a live Squads editor (correct 4-4-2,
5-star squad, no freezes).
Key findings this session:
- userMassInfo MUST stay {} (any content desyncs the massinfo parser
0x180174630 -> tokenizer busy-loop freeze). Deliver the squad via
GET /squad/0 (fetched on Squads-tab entry) instead.
- Player cards render generic because the card view-model (0x1800d7920)
reads identity/rating/face from a resolved record at item+0x10, filled
by a lookup (0x18011cca0) in the FUT item-definition std::map at
CardsDb+0x160c0 -- which is EMPTY offline -> default blank record.
- Version advertising (itemDbVersion/checkServerDbVersion) is proven inert
(JSON fields routed to the skip handler). Owned items don't auto-trigger
a definition fetch. In-place map overwrite is dead (map stays empty).
- Definition-serving endpoints (item/resource, defid, item?idList) built +
ready; the fetch trigger lives in the packed FIFA17.exe.
New: docs/CARD_SYSTEM.md (findings + ordered next-steps plan for real
player cards: patch-POC, dbdata extractor, drive FIFA17.exe fetch, or
live-memory store injection). Plus tools: fut_seed.py (squad ladder +
definition serving), fifadrive.sh, vgamepad.py, and the login-RE toolset.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PN5bmpDVQR1aXgefyWAt7o
Emulates FIFA 17's full online + Ultimate Team stack against an offline,
clean-room backend (no EA servers). Proven end-to-end 2026-08-01:
Origin login -> Blaze login -> device-trust -> the FUT hub.
Package:
- tools/openfut-fut.sh one-command orchestrator (start/stop/status/restart)
- tools/root_arm.sh idempotent host arm (sysctls, DNAT, /etc/hosts easw)
- tools/{lsx_responder_v2,blaze_responder_v3b,roster_server,utas_server,autopatch}.py
the 5 servers (Origin LSX :4216, Blaze :42127/42130/42131, roster :8081,
FUT/UTAS :8099) + heat2.py (Fire2/Heat2 TDF codec)
- FUT-RUNBOOK.md runbook + gate-ladder troubleshooting
- docs/, tools/login_dump/*.md the reverse-engineering write-ups
All findings are clean-room, from binaries we own; nothing from any leak.
The wire protocol maps 1:1 to FIFA 23.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PN5bmpDVQR1aXgefyWAt7o
Adds the app-centric FLE bridge direction (direction.md) replacing the
Blaze-backend route as primary plan, plus a corrected foundational test
procedure and snapshot/diff/apply tooling for reverse-engineering FLE's
Freeze Lineup write mechanism. Also picks up prior untracked research docs
(status-review, fut-integration-options, fifa23-startup-flow, track-c) and
existing capture/export tooling that hadn't been committed yet.
Registers the new fifa-blaze workspace as a submodule.
Milestone 1: capture stub that logs every Blaze packet to JSONL.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The broad OUTPUT 443→8443 redirect caught all HTTPS traffic system-wide.
Scoping to -d 127.0.0.1 means only EA domains (resolved to 127.0.0.1 via
/etc/hosts) are redirected; all real internet traffic is unaffected.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-06-26 09:13:00 -07:00
1010 changed files with 685113 additions and 8 deletions
This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
> ⚠️ **Stale (FIFA 23).** This file's status and targets predate the FIFA 17 pivot. Prefer [`docs/PROJECT_STATE.md`](./docs/PROJECT_STATE.md) (canonical). The working target is **FIFA 17**; the canonical server is `fifa17-recon/docker/fifa17-python` (`docker compose up -d`). `openfut-bridge` (FIFA 23) is superseded; `openfut-core` remains the shared backend.
## Repository Layout
This is a monorepo containing three independent Rust crates as git submodules:
> ⚠️ **Status — see [`docs/PROJECT_STATE.md`](./docs/PROJECT_STATE.md) (canonical).** The working, actively-developed target is **FIFA 17**, not FIFA 23. Everything below this banner describes the **superseded FIFA 23 `bridge` lineage** and is kept for historical context.
>
> **Run the server (canonical):** `cd fifa17-recon/docker/fifa17-python && docker compose up -d` — see [`fifa17-recon/FUT-RUNBOOK.md`](./fifa17-recon/FUT-RUNBOOK.md). `openfut-core` is the shared offline backend (still used by the FIFA 17 path); `openfut-bridge` is the retired FIFA 23 integration.
**Offline Ultimate Team — like SPT, but for FIFA 23.**
OpenFUT replaces EA's retired FUT servers with a fully offline, single-player backend. You own FIFA 23 legitimately. You just want to keep playing after EA shut down the servers.
"title":"Missing Authentication on All Endpoints",
"description":"No authentication or authorization checks on any API endpoint. The system uses single-profile design with get_active_profile() returning the first row (LIMIT 1) without any token validation, session management, or per-user isolation. In a networked context, any HTTP client can access all endpoints without credentials.",
"impact":"Complete compromise of data confidentiality and integrity. Any attacker can view, modify, or delete all user data without authentication.",
"exploitPath":"curl http://127.0.0.1:8080/clubs - accesses club data without any auth headers or tokens",
"recommendation":"Implement stateless JWT tokens or session-based authentication. Add middleware to validate tokens on all endpoints. Implement per-user authorization checks in services."
},
{
"severity":"critical",
"category":"injection",
"file":"openfut-core/src/routes/auth.rs",
"line":87,
"cwe":"CWE-89",
"title":"SQL Injection via String Interpolation",
"description":"SQL table names are interpolated using string formatting: sqlx::query(&format!(\"DELETE FROM {table}\")). Although currently hardcoded in a loop, this violates parameterized query principles and creates a risk if the table list ever becomes user-controlled or the pattern is copied elsewhere.",
"impact":"Potential remote code execution via database manipulation. If extended to user input, attackers could modify arbitrary tables or drop the database.",
"exploitPath":"Currently mitigated by hardcoded table names, but the pattern is dangerous and violates secure coding practices.",
"recommendation":"Use SQLx's dynamic query builders or identifier types that properly escape table/column names. Replace format! string interpolation with sqlx::query_builder for dynamic identifiers."
},
{
"severity":"high",
"category":"configuration",
"file":"openfut-bridge/src/proxy.rs",
"line":44,
"cwe":"CWE-295",
"title":"TLS Certificate Validation Disabled",
"description":"HTTP client explicitly disables TLS certificate validation: .danger_accept_invalid_certs(true). This bypasses all certificate pinning, expiration, and hostname verification, making the bridge vulnerable to man-in-the-middle attacks.",
"impact":"Attacker positioned between bridge and upstream can intercept, modify, or read all traffic. Compromises confidentiality and integrity of requests to Core and external services.",
"exploitPath":"MITM attack between openfut-bridge and openfut-core or upstream services. ARP spoofing on localhost subnet would redirect traffic.",
"recommendation":"Remove .danger_accept_invalid_certs(true) in production. If testing requires it, gate behind a development-only environment variable with strong warning. Use proper certificate management (CA bundles, cert pinning)."
},
{
"severity":"high",
"category":"dos",
"file":"openfut-core/src/services/season.rs",
"line":23,
"cwe":"CWE-248",
"title":"Unguarded expect() Causes Denial of Service",
"description":"Multiple unchecked expect() calls that will panic and crash the server if database queries fail or return unexpected results: Ok(fetch(pool, profile_id).await?.expect(\"just inserted\"))",
"impact":"Denial of service. A single database inconsistency or race condition crashes the entire server, making the application unavailable.",
"exploitPath":"Trigger race conditions during concurrent requests (e.g., rapid profile deletion + season fetch). Database corruption or migration failure crashes the service immediately.",
"recommendation":"Replace expect() with proper error handling (Result types, error logging, graceful degradation). Handle database query failures without panicking. Add integration tests for race conditions."
},
{
"severity":"high",
"category":"dos",
"file":"openfut-core/src/services/season.rs",
"line":69,
"cwe":"CWE-248",
"title":"Unguarded expect() in season fetch",
"description":"let season = fetch(pool, profile_id).await?.expect(\"season must exist\"); Panics if season is not found.",
"impact":"Server crash on missing or deleted season records.",
"exploitPath":"Delete a season via concurrent requests, then call /seasons endpoint. Server panics.",
"recommendation":"Return proper error (AppError::NotFound) instead of panicking."
},
{
"severity":"high",
"category":"dos",
"file":"openfut-core/src/services/season.rs",
"line":144,
"cwe":"CWE-248",
"title":"Unguarded expect() in season update",
"description":"let updated = fetch(pool, profile_id).await?.expect(\"season must exist\");",
"impact":"Server crash on concurrent season modifications.",
"exploitPath":"Rapid concurrent season updates that fail race conditions.",
"recommendation":"Handle missing records gracefully."
},
{
"severity":"high",
"category":"cors",
"file":"openfut-core/src/app.rs",
"line":257,
"cwe":"CWE-346",
"title":"Permissive CORS Configuration Allows All Origins",
"description":".layer(CorsLayer::permissive()) enables CORS for all origins (*), methods, and headers. Any website can make cross-origin requests to the API and access/modify data.",
"impact":"Cross-site request forgery (CSRF) attacks. Malicious websites can issue API requests on behalf of users. Data exfiltration via JavaScript from any origin.",
"exploitPath":"Attacker website:\n <img src=\"http://127.0.0.1:8080/clubs\" />\n Fetch API calls to delete profiles, modify squads, etc.",
"recommendation":"Restrict CORS to specific origins (e.g., localhost:3000 for web UI, or the game process if exposed). Use CorsLayer::very_restrictive() as default and explicitly allowlist origins."
},
{
"severity":"high",
"category":"dos",
"file":"openfut-bridge/src/proxy.rs",
"line":47,
"cwe":"CWE-248",
"title":"HTTP Client Construction Panic",
"description":".expect(\"failed to build HTTP client\") will panic if the HTTP client fails to initialize, crashing the entire proxy service on startup.",
"impact":"Service unavailability. Bridge cannot start if HTTP client configuration is invalid.",
"exploitPath":"Invalid system configuration or missing TLS libraries causes HTTP client build to fail, crashing bridge during startup.",
"recommendation":"Return Result<ProxyState, Error> from new() and handle construction errors. Use anyhow::Context for better error messages."
"description":"JSON parsing errors are returned directly to clients: format!(\"json parse error: {e}\"). Exposes serde_json parser internals and syntax details useful for crafting attacks.",
"impact":"Information disclosure. Attackers learn the JSON parser implementation and can tailor payloads to bypass validation or find parser-specific quirks.",
"exploitPath":"Send malformed JSON to any endpoint. Response includes parser error details (e.g., 'expected `,` at line 2 col 5') that aid in crafting exploits.",
"recommendation":"Return generic error message to clients: 'invalid request format'. Log detailed errors internally with tracing for debugging."
},
{
"severity":"medium",
"category":"information-disclosure",
"file":"openfut-core/src/error.rs",
"line":40,
"cwe":"CWE-215",
"title":"Database Errors Logged with Full Details",
"description":"Database errors are logged with full SQL/query details: tracing::error!(\"Database error: {e}\"). If logs are exposed or compromised, schema, query patterns, and data structure are revealed.",
"impact":"Information disclosure in logs. Compromised log files expose database schema and query logic useful for SQL injection or data exfiltration planning.",
"exploitPath":"Access server logs (via log aggregation service, file access, etc.) and extract database schema and query patterns.",
"recommendation":"Log only error type and ID to clients. Sanitize logs before exporting. Use structured logging with field masking for queries."
},
{
"severity":"medium",
"category":"input-validation",
"file":"openfut-core/src/routes/auth.rs",
"line":17,
"cwe":"CWE-1025",
"title":"Hardcoded Default Credentials",
"description":"Default username 'Player 1' is hardcoded with no unique identifier enforcement. Multiple profiles can be created with identical usernames, and weak defaults are used.",
"impact":"Weak account creation, potential for account confusion or conflicts. No strong identity guarantees.",
"exploitPath":"Multiple users create profiles with default 'Player 1' username. No way to distinguish profiles programmatically.",
"recommendation":"Require explicit username on profile creation. Use UUIDs as primary identifiers. Validate username uniqueness and minimum length."
},
{
"severity":"medium",
"category":"input-validation",
"file":"openfut-core/src/services/",
"line":0,
"cwe":"CWE-400",
"title":"Missing Input Length Validation",
"description":"No maximum length checks on string fields (usernames, club names, squad names, etc.). Large inputs can cause database bloat, memory exhaustion, or DoS.",
"impact":"Denial of service via large payloads. Database bloat. Memory exhaustion. While DefaultBodyLimit::max(256KB) provides some protection, field-level validation is missing.",
"recommendation":"Add input validation for all user-submitted strings. Set maximum lengths (e.g., username: 50 chars, club name: 100 chars). Validate at route handler level."
},
{
"severity":"medium",
"category":"configuration",
"file":"openfut-core/src/db.rs",
"line":13,
"cwe":"CWE-315",
"title":"Unencrypted SQLite Database on Disk",
"description":"SQLite database file (openfut.db) is stored unencrypted on disk. All user data, profiles, squads, cards, etc., are readable by anyone with filesystem access.",
"impact":"Data breach if server filesystem is compromised. No protection against:local file access, stolen backups, forensic recovery.",
"exploitPath":"Attacker gains filesystem access (compromised server, stolen disk). Reads openfut.db directly. All game data is readable without authentication.",
"recommendation":"Use SQLite encryption (e.g., sqlcipher crate) or migrate to PostgreSQL with TLS. Implement file-level encryption. Use restrictive filesystem permissions (0600)."
},
{
"severity":"medium",
"category":"rate-limiting",
"file":"openfut-core/src/app.rs",
"line":0,
"cwe":"CWE-770",
"title":"No Rate Limiting on Endpoints",
"description":"No per-IP or per-user rate limiting. Endpoints like POST /auth/reset can be called repeatedly without restriction, allowing attackers to repeatedly wipe all data.",
"impact":"Denial of service and data destruction. Attacker can spam /auth/reset to destroy user data or exhaust server resources.",
"exploitPath":"for i in 1..1000: POST /auth/reset with confirm='reset'. All data wiped repeatedly.",
"recommendation":"Implement rate limiting middleware using tower_governor or similar. Add per-IP limits (e.g., 10 requests/min) and per-endpoint limits. Use exponential backoff."
},
{
"severity":"low",
"category":"audit-logging",
"file":"openfut-core/src/services/",
"line":0,
"cwe":"CWE-778",
"title":"Missing Audit Logging",
"description":"No audit trail of user actions (profile creation, data deletion, squad modifications). Cannot detect unauthorized access, data tampering, or compliance violations.",
"impact":"Incident response and forensics are impossible. Cannot determine who did what and when. Compliance risks (GDPR, etc.).",
"exploitPath":"Attacker deletes all profiles, modifies squads. No audit log shows what happened or who did it.",
"recommendation":"Add audit logging for all data mutations. Log: timestamp, user (profile) ID, action, resource affected, before/after state. Store in separate immutable table."
"description":"openfut-bridge uses reqwest 0.11 (latest is 0.12) and rustls 0.21 (latest is 0.23). Intentional for version matching, but creates a larger surface area for known CVEs.",
"impact":"Potential vulnerabilities in older dependencies. Delayed access to security patches.",
"exploitPath":"Known CVE in reqwest 0.11 or rustls 0.21 could be exploited. Combined with danger_accept_invalid_certs, TLS bypass becomes easier.",
"recommendation":"Upgrade dependencies to latest versions when possible. Monitor CVE databases (CVE, RustSec) for the versions in use. Pin versions and set up automated dependency updates."
},
{
"severity":"low",
"category":"error-handling",
"file":"openfut-core/src/app.rs",
"line":256,
"cwe":"CWE-248",
"title":"Body Size Limit Without Per-Field Validation",
"description":"DefaultBodyLimit::max(256KB) limits the entire request body, but individual fields are not validated. A single large field can consume most of the limit.",
"impact":"Mild DoS. Large field values cause database bloat. Not a critical issue due to body limit, but field-level validation would be better.",
"exploitPath":"POST /auth/local with 250KB club_name field. Database receives bloated data.",
"recommendation":"Add per-field validation in addition to body limits. Validate and sanitize fields before database insertion."
}
],
"riskScore":82,
"riskCategory":"CRITICAL",
"riskSummary":"OpenFUT has critical security issues that would make it unsafe for production or networked deployment. The most severe are the complete absence of authentication/authorization and the SQL injection pattern in the auth.rs module. The system is designed as single-player (single-profile) with no multi-tenant isolation, which is dangerous if exposed to the network.",
"recommendations":[
{
"priority":"CRITICAL",
"area":"Authentication & Authorization",
"recommendation":"Implement JWT-based or session-based authentication on all endpoints. Add middleware to validate auth tokens on every request. Implement per-profile authorization checks. Currently any HTTP client can access all endpoints.",
"effort":"High",
"impact":"Blocks all data breaches from unauthenticated access"
},
{
"priority":"CRITICAL",
"area":"SQL Injection Prevention",
"recommendation":"Replace sqlx::query(&format!(...)) in auth.rs:87 with proper parameterized identifiers. Use sqlx::query_builder for dynamic table/column names instead of string interpolation.",
"effort":"Low",
"impact":"Prevents SQL injection even if pattern is copied to user input"
},
{
"priority":"HIGH",
"area":"TLS & Transport Security",
"recommendation":"Remove .danger_accept_invalid_certs(true) from proxy.rs:44. If development requires it, gate behind an environment variable (e.g., DEV_SKIP_TLS_VERIFICATION) with strong warnings in logs.",
"effort":"Low",
"impact":"Prevents MITM attacks on bridge-to-core communication"
},
{
"priority":"HIGH",
"area":"Error Handling",
"recommendation":"Replace all expect() calls with proper Result handling. Use anyhow::Context or custom error types. Add logging for debugging but return generic errors to clients.",
"effort":"Medium",
"impact":"Prevents DoS via server panics"
},
{
"priority":"HIGH",
"area":"CORS",
"recommendation":"Replace CorsLayer::permissive() with CorsLayer::very_restrictive() or explicit allowlist. For single-player use, restrict to localhost and the game process only.",
"effort":"Low",
"impact":"Prevents CSRF and cross-origin attacks"
},
{
"priority":"HIGH",
"area":"Rate Limiting",
"recommendation":"Add per-IP rate limiting using tower_governor or similar. Implement limits on destructive endpoints (e.g., POST /auth/reset: 1 request per hour per IP).",
"effort":"Medium",
"impact":"Prevents DoS and repeated data destruction"
},
{
"priority":"MEDIUM",
"area":"Input Validation",
"recommendation":"Add maximum length validation for all string fields (username, club_name, squad_name, etc.). Enforce at route handler level. Example: username max 50 chars, club_name max 100 chars.",
"effort":"Medium",
"impact":"Prevents database bloat and data validation failures"
},
{
"priority":"MEDIUM",
"area":"Data Encryption",
"recommendation":"Use SQLite encryption (sqlcipher) or migrate to PostgreSQL with TLS. Set restrictive filesystem permissions (0600) on openfut.db.",
"effort":"High",
"impact":"Protects data at rest from filesystem access"
},
{
"priority":"MEDIUM",
"area":"Error Message Handling",
"recommendation":"Return generic error messages to clients. Log detailed errors internally. Example: client sees 'invalid request', server logs 'JSON parse error: expected `,` at line 2'.",
"effort":"Low",
"impact":"Reduces information disclosure"
},
{
"priority":"MEDIUM",
"area":"Audit Logging",
"recommendation":"Add audit trail for all data mutations (create, update, delete). Log timestamp, profile ID, action, resource, and before/after state. Store in immutable audit_log table.",
"effort":"Medium",
"impact":"Enables incident response and forensics"
},
{
"priority":"LOW",
"area":"Dependency Management",
"recommendation":"Upgrade reqwest to 0.12 and rustls to 0.23 when possible. Set up Dependabot or RustSec monitoring for CVEs. Regularly audit dependencies.",
"effort":"Low",
"impact":"Reduces attack surface from known CVEs"
},
{
"priority":"LOW",
"area":"Default Values",
"recommendation":"Remove hardcoded default username 'Player 1'. Require explicit username on profile creation. Use UUIDs for profile identification.",
"effort":"Low",
"impact":"Improves account identity and prevents confusion"
}
],
"securityDesignNotes":{
"intendedUse":"OpenFUT is designed for single-player offline use. Single-profile design is intentional for local FIFA 23 emulation.",
"deploymentContext":"Localhost only (127.0.0.1:8080). Not intended for networked or multi-user deployment.",
"implicationForSecurity":"Many security issues (no auth, permissive CORS) are acceptable for localhost-only use. However, the code structure lacks security boundaries, so if ever exposed to the network, it would be completely unsecured. Recommend adding security gates now rather than retrofitting later.",
"suggestedDefensiveApproach":"Even for single-player use, add security layers (basic auth, CORS restrictions, rate limiting) to prevent accidental misuse if deployed in an unsafe context."
},
"positiveFindingsAndStrengths":[
"✓ SQLx is used throughout with parameterized queries (except auth.rs:87)",
"✓ Foreign key constraints are enforced in SQLite",
"✓ UUIDs are used for entity IDs instead of sequential IDs (reduces enumeration attacks)",
"✓ Request body size is limited to 256KB (prevents large payload DoS)",
"✓ Concurrency is limited to 256 concurrent requests",
"✓ Sensitive tokens (X-UT-SID, X-UT-PHISHING-TOKEN) are stripped from captures",
"✓ Logging is structured using tracing crate (good for audit trails)",
"Add fuzz testing for JSON parsing to find edge cases"
],
"complianceNotes":{
"gdpr":"No explicit data handling policy. If user data is processed, GDPR requires consent, data retention limits, and audit trails. Not currently implemented.",
"dataProtection":"Unencrypted database at rest violates most data protection frameworks.",
"logging":"Audit logging is missing, violating compliance requirements."
File diff suppressed because one or more lines are too long
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
funman300