battle_assassinate times out driving toward Shadow's divine

[mtklein]

Found while triaging the five suites #236 turned red, and confirmed to be neither that feature's nor the triage's:

FAIL: timeout after 30000 frames driving toward Shadow chips a gauge to Broken
      and his divine kills it

It fails identically on a control ROM with #236's four runtime hooks NOP'd in place (layout byte-identical), and its log contains zero fleeBattle calls, so the lib change that fixed battle_slots is not implicated either.

Not yet diagnosed. Worth checking against this cycle's recurring fault first — a precondition the run hopes for rather than reaches — but that is a hypothesis to test, not a conclusion.

[mtklein]

Drive fixed on main in 0d01935 (wt/harness-faults): the timeout was the fight being WON with nothing Broken while the drive idled on the spoils (bench Defend misread state $27; Interceptor's counters killed bodies outright; the target was hoped PIERCE-weak). Bench Defends via $27, target steered, Interceptor on the ledger, spoils pressed out, up to six encounters, break detection from the ROM's own OT6_BROKEN_TICKS write.

The finding that keeps this open is a game one: OT6's breaking hit is doubled (Ot6BrokenDmg) and Ot6Assassinate resolves at ChooseTarget, before that hit's chip, so Shadow's divine can only take a body that is still standing AFTER its own break -- ~280 hp against the back-row Imperial -- and every PIERCE-weak body camp_escaped's pool deals (CrassHoppr, 243) dies on its own break. Four fights running proved it; the only divine in five fights was a berserked Shadow's #236 pip dump breaking and killing inside one frame. Boost buys swings, not damage, so no pip count moves the floor; the bag has no weaker dagger. The instrument now asks for two PIERCE-weak 280+ bodies and this pool cannot deal them: 0/8 at the precondition. Owner's call: (A) accept the mechanic as designed and move the instrument to a fixture whose pool has 300+ PIERCE-weak bodies, or (B) let the divine fire on the breaking hit itself, so 'break it and it dies' is Shadow's thing (still once per battle). Evidence: build/attempts/wt/harness-faults/lab/harness-faults/repro/assassinate_fixed_s0.log and frames/.

[mtklein]

Owner ruling (2026-09-21): option B -- 'break it and it dies' is Shadow's thing. When Shadow's own hit breaks a non-boss body, the divine kills it on that hit, still once per battle and never a boss. An Octopath-style twist for an FF6 character, which is exactly the kind of thing this project is for. In progress on wt/assassinate-break, with battle_assassinate reworked to prove it on camp_escaped, where it can now fire.

[mtklein]

Merged as 52e2025 (branch wt/assassinate-break). Shadow's breaking hit is his divine -- break it and it dies -- plus the already-Broken case kept. One gate/kill Ot6AssassinateGate, reached from Ot6HitJoin (after the class chip, so the hit that empties the last shield reads its own break) and the existing ChooseTarget seam. Once per battle, never a boss.

Evidence (build/attempts/wt/assassinate-break): measured order inside one action -- break store (x=Shadow) seq1 -> Death mark seq2 -> ApplyDmg HP write seq3 -> UpdateStatus Death seq4. Suite battle_assassinate PASS (frame 6773); 8/8 seed sweep, distinct first-battle keys $be=18/34/50/6C/88/A4/C0/DC; negative control (Ot6HitJoin jsr NOP'd, layout identical) fails 'got seam, want Ot6HitJoin'. nomp_distinct.ok passes (the control ROM moves, as a game rule must -- not gated under OT6_MP_COSTS).