test.lua is the gate -- it fails on what is broken. This is the map: it
reports what is merely suspicious, and (the actual point) it REMEMBERS.
32 jokers score ~0 against all 150: they give nothing but engine roles
and want nothing, so the advisor has nothing to say about them. For most
that is the correct answer, but I have now worked that question out from
scratch twice, and the reasoning evaporated both times. VERDICTS records
the conclusion for each one, so an audit only surfaces what is new.
It preserves the traps: Steel Joker and Glass Joker look exactly like
Blue Joker and Abstract Joker did, but they want Steel/Glass cards
specifically, and the only enhance_gen givers make Gold and Stone -- so
"fixing" them would invent a synergy that does not exist. That is written
down now instead of being rediscovered.
It also preserves the leads not yet chased (Ceremonial Dagger's
positional destroy, Erosion vs the card-adders, Burglar + Ramen and Green
Joker, Chaos + Flash Card), marked candidate because they are unverified.
A newly-silent joker lands in "never looked at" and fails the audit until
someone decides. AUDIT.md is committed so the map is readable without
running anything.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
funman300