Files
BFM-decomp/docs
Drew T 89162fd6b8 feat(phase-25): task A giant batch 1 — 2 more giants ×134 (func_80135480, func_80163EC8) + §44 levers; 4 Fable5 seeds
- cheap-Opus batch over the 6 frontier giants: 3 banked ×134 this session (func_80166994 §43
  committed earlier; + func_80135480 258-ins block-scoped-pointer-split; func_80163EC8 234-ins
  cross-jump-duplicated-tail, 1 benign $v0 pin). 266 siblings byte-identical, 0 failed
- R22 clean-fleet 136/136; instr-weighted 57.2->57.7%, distinct-code 28.2->29.3%, dedup 1813/0
- cookbook §44 — 5 reusable levers: §43-extension (K&R s16 also covers callee-stash sign-extend),
  pointer-var-decl (avoid &sym CSE-hoist), block-scoped-pointer-split, cross-jump-duplicated-tail
  (cracks a §31-D1/D2-"permuter-only" class), + the intrinsic-wall taxonomy (what cheap-Opus can't)
- R14: §43 does NOT universally transfer (only 1 of 6 giants was K&R-s16; each is its own class);
  giant #1's prior "MATCH" note was stale/false (verify vs bytes)
- 4 giants -> pin-free/structural Fable5+permuter SEEDS (func_80133CD4 flagship §37 allocno-tie,
  func_8014D820 RC-6 pin-free, func_801670E4 scheduling, func_8016CBC0 coalescing); escalation
  = permuter-first then Fable5 §34, queued in the CURRENT_PHASE resume block
- SESSION CHECKPOINT: phase 25 OPEN; fresh session resumes the giant-seed escalation round
2026-07-11 02:27:23 -06:00
..