mirror/BFM-decomp
1
0
Fork 0
mirror of https://github.com/Druthulu/BFM-decomp synced 2026-09-26 13:33:34 -04:00
Code Issues Packages Projects Releases Wiki Activity

Commit Graph

  • 99b65ef94a docs(phase-32): T4b — func_80185810 35 -> 13 NEAR (Fable, 471k tokens): 3 of 4 windows closed (P_TAG bitfield qty priorities, sched1 flush_pending_lists load order, hard-reg dests not boosted); residual = the cl second-set fence; backlog row + draft + report kept Drew T 2026-09-05 19:09:36 -06:00
  • 066a2f1cce docs(phase-32): cookbook §501-H — large constants are unboosted floaters; the prologue weave is decided by which ready-list stalls eat them (func_800CD92C, zero pins); T4b row (8 banked) Drew T 2026-09-05 18:55:38 -06:00
  • 511afcf75c docs(phase-32): T4b (8) ledger — func_800CD92C banked (md_MAIN_009 100% C), backlog re-rendered, the Fable draft + report + verdict kept; queue updated Drew T 2026-09-05 18:55:02 -06:00
  • 4a9a67c59e feat(phase-32): T4b (8) — md_MAIN_009: func_800CD92C (247 ins) BANKED byte-identical d270f695 — the prologue-weave NEAR falls with ZERO pins: m24 wins $t1 by flow ref count (the P_TAG bitfield's foldable re-mask), and the tpage constants float to the top because a multi-set tag load (bitfield RMW shape) fills the RMW latency stalls instead of being birthing-boosted (Fable agent); md_MAIN_009 is 100% C Drew T 2026-09-05 18:54:50 -06:00
  • d7561f2680 docs(phase-32): cookbook §501-G — three passes, three dials (sched1 birthing boost / a phantom USE nop in sched2's model / a local-alloc tie broken by the P_TAG bitfield's flow ref count; func_800CF3E8); T4b row (7 banked, 1 near-proved) Drew T 2026-09-05 18:43:53 -06:00
  • 5d8b8478fa docs(phase-32): T4b (7) ledger — func_800CF3E8 banked (md_MAIN_003 100% C), backlog re-rendered, the Fable draft + report + verdict kept; queue updated Drew T 2026-09-05 18:43:11 -06:00
  • cb871087ec feat(phase-32): T4b (7) — md_MAIN_003: func_800CF3E8 (469 ins) BANKED byte-identical — the 27-off alias basin falls: a second live set of the loaded tag kills sched1's birthing boost, the phantom ot-USE that filled the load-delay slot only in the scheduler's model is removed, and the OT link spelled as libgpu's P_TAG bitfield lifts m24's flow ref count so it wins $a2 in local-alloc (Fable agent) Drew T 2026-09-05 18:43:03 -06:00
  • 3c6d20991f docs(phase-32): T4b — func_80032A74 stays NEAR 1 (Fable, 402k tokens): the 0x48 slot is a GHOST pseudo (combine.c:2306-2313), the caller-save-area hypothesis REFUTED with citations, the only memory-value ghost species is the lh SIGN_EXTEND split; pin annotated, backlog row, draft + report kept Drew T 2026-09-05 18:35:27 -06:00
  • ce4cdaf750 docs(phase-32): cookbook §501-F — split the CSE quantity when loop.c proves the hoist split unreachable (hard-reg dividend copy + hoisted sign; func_801834A4); T4b row (6 banked) Drew T 2026-09-05 18:26:11 -06:00
  • 1081b02b66 docs(phase-32): T4b (6) ledger — func_801834A4 unpinned (2 walls remain), backlog re-rendered, the Fable draft + report + verdict kept; queue updated Drew T 2026-09-05 18:25:29 -06:00
  • ff86e25729 feat(phase-32): T4b (6) — ov_SC03_105: func_801834A4 (106 ins) BANKED byte-identical — a pinned WALL falls: a CSE-quantity split hoists only the division's sign correction (sign = half >> 31 at the loop top; the divide reads a hard-register copy so its mult is not a movable; expand_divmod's sra is CSE'd into sign) (Fable agent) Drew T 2026-09-05 18:25:13 -06:00
  • 78613864e4 docs(phase-32): cookbook §501-E — a register pin forbids that register to every retried allocno (regs_ever_live → bad_spill_regs → retry_global_alloc; func_80020DA4); T4b row (5 banked) Drew T 2026-09-05 18:16:51 -06:00
  • 9564314095 docs(phase-32): T4b (5) ledger — func_80020DA4 unpinned (3 walls remain), backlog 11 open, the Fable draft + report + verdict kept; resume queue (3-at-a-time) tracked Drew T 2026-09-05 18:16:50 -06:00
  • b47bc2ea0b feat(phase-32): T4b (5) — main: func_80020DA4 (100 ins) BANKED byte-identical 143dbb89 via gate_main — a pinned WALL falls: the $6 pin on e0 FORBADE $a2 at reload's retry_global_alloc (regs_ever_live seeds bad_spill_regs), so the product took $t0; unpinned + two zero-byte launders steer local-alloc's qty_compare to the target's allocation (Fable agent) Drew T 2026-09-05 18:16:01 -06:00
  • ad62b693a1 docs(phase-32): cookbook §501-D — one value-returning call makes the call a birthing insn (reg_n_sets[$v0]==1 → sched1 max priority → the def above the jal); a second $v0 set via the callee's real return type (func_8017DF28); T4b row (4 banked) Drew T 2026-09-05 13:29:17 -06:00
  • 32ff170bf3 docs(phase-32): T4b (4) ledger — func_8017DF28 unpinned (4 walls remain), backlog re-rendered, the Fable draft + report + verdict kept; bank.sh takes SPLIT for _jr_ TUs and prints rtu's tail when no verdict line appears Drew T 2026-09-05 13:28:35 -06:00
  • b31084300f feat(phase-32): T4b (4) — ov_SC06_022: func_8017DF28 (119 ins) BANKED byte-identical — a pinned WALL falls: one value-returning call made reg_n_sets[$v0]==1 and sched1's birthing boost scheduled it above the &mtx def (reorg then filled the jal slot); a second $v0 set via RotMatrixY's real pointer-returning type puts the def after the call and reorg fills the bnez slot (Fable agent) Drew T 2026-09-05 13:28:24 -06:00
  • 1f024f8e4b docs(phase-32): cookbook §501-C — the dying-input suggestion vs the birthing boost (func_800CD674); backlog 13 open; bank.sh DRAFT_DIR + R43 hardening; T4b row (3 banked) Drew T 2026-09-05 13:26:52 -06:00
  • b13e824f76 feat(phase-32): T4b (3) — md_MAIN_009: func_800CD674 (174 ins) BANKED byte-identical d270f695 — the $a3<->$t1 REGALLOC-PERM plateau falls: prim 4 gets a fresh single-death masked-p (local-alloc's dying-input suggestion needs a pseudo BORN at the and) and prim 3's copy gets a trailing zero-byte asm set that defeats sched1's birthing boost (Fable agent). CORRECTION: commits commit:3962 and commit:3963 claimed this bank before it happened — bank.sh had been called without the function name, then with the wrong draft dir; both times it refused or built the unchanged tree Drew T 2026-09-05 13:25:57 -06:00
  • 3a9c946d8d docs(phase-32): T4b (3) ledger — backlog re-rendered after the real func_800CD674 bank; bank.sh refuses an empty function list and propagates a failed commit (R43 — it had built the unchanged tree and exited 0) Drew T 2026-09-05 13:25:30 -06:00
  • 779b55a497 docs(phase-32): T4b (3) ledger — func_800CD674 banked (backlog 13 open), the Fable draft + report + verdict kept; bank.sh's verbatim check now refuses only instruction-bearing asm (zero-byte fences/launders pass) Drew T 2026-09-05 13:24:05 -06:00
  • 2d0957ffb1 docs(phase-32): cookbook §501-B — do not merge case tails in C (cross-jump merges post-alloc; a C-level merge deletes the reference that decides allocno order — func_80039DEC); T4b row Drew T 2026-09-05 13:21:10 -06:00
  • d913bca9b7 docs(phase-32): T4b (2) ledger — func_80039DEC unpinned (5 walls remain), backlog re-rendered, the Fable draft + report + verdict kept (R20) Drew T 2026-09-05 13:20:36 -06:00
  • e8c2b6baa0 feat(phase-32): T4b (2) — main: func_80039DEC (74 ins) BANKED byte-identical 143dbb89 via gate_main — a pinned WALL falls: the natural switch with DUPLICATED case tails (cross-jump merges them post-reload), so a2's raw copy keeps 4 refs and wins $a3 (Fable agent) Drew T 2026-09-05 13:20:25 -06:00
  • ec5572c5fb docs(phase-32): cookbook §501 — a lever that measures worse may be a cascade (func_800391D4: the explicit promotion flipped loop.c's insn_count threshold; pads 7→9 closed it); accelerators (9); T4b row Drew T 2026-09-05 13:16:50 -06:00
  • 177f4a7dfd docs(phase-32): T4b (1) ledger — func_800391D4 unpinned (config/wave_exclude.txt 6 kept / 1 dropped by exclude_audit), backlog re-rendered (15 open), the Fable draft + report + verdict kept (R20) Drew T 2026-09-05 13:15:53 -06:00
  • 1e54e88a81 feat(phase-32): T4b (1) — main: func_800391D4 (75 ins) BANKED byte-identical 143dbb89 via gate_main — a pinned WALL falls: the sign-extend must be preheader SOURCE code, and the loop.c insn_count knife-edge re-padded (Fable agent) Drew T 2026-09-05 13:15:20 -06:00
  • f85b589c3f docs(phase-32): 🛑 checkpoint — T4b in flight: 15 Fable agents launched ~13:05 MDT from .run/P32/t5x (BRIEF + packs + the hand-pass NOTES); recovery route via agent_verdicts over this session's a*.output; T5 follows T4b Drew T 2026-09-05 12:59:41 -06:00
  • 34268e3b8e docs(phase-32): T4b hand pass — all 15 rows hand-analysed against the compiler; 22 spellings measured (all regressions), 0 banks; every blocker refined to a named mechanism + untried lever in .run/P32/t4b/<fn>/NOTES.md; Fable brief + targets + packs staged Drew T 2026-09-05 12:55:49 -06:00
  • e392290513 chore(phase-32): T4b plumbing — src/800_c.c: func_80039DEC decl → no-proto; D_80073140 externs (×4) → [][1] with (s32*) casts at the 3 uses — byte-neutral (in-tree clean extract+build 143dbb89) Drew T 2026-09-05 12:36:39 -06:00
  • 90acd6902f docs(phase-32): T4 DONE — the walls' FINAL ledger: 7 pinned rows re-probed in TU context (3 CC1 FAILs were plumbing, re-probed in a sandbox TU), 1 PROVED + 6 CANDIDATE, no verdict changed; pins annotated, backlog rows for all 7 (R62 paths fixed), cookbook §500-I, decision-log Drew T 2026-09-05 12:11:10 -06:00
  • a373a000d1 docs(phase-32): T3 CLOSED — 39 banks in the pass (29 this session), census 15 stubs / 3,758 ins (7 pinned walls + 8 NEAR), fleet R22 218/218 ×3 today, fleet instr 100.0% / distinct 99.9% / fn 100.00%; §500-H; 🛑 block → T4 Drew T 2026-09-05 12:01:03 -06:00
  • 9471d8c01e docs(phase-32): cookbook §500-G — the func_8001BC6C crack (permuter waypoint minus its width mutation = the §47 birth lever, spelled well-defined), the func_800CD674 plateau, the masked_diff-probe/Makefile collision; harvest_notes; index regenerated Drew T 2026-09-05 11:39:04 -06:00
  • 563a211448 build: exclude dotfiles from the C source find (-not -name '.*') — a tool's live probe in src/ broke gate_main's clean rebuild (P32 S83) Drew T 2026-09-05 11:37:53 -06:00
  • 836181a6c2 feat(phase-32): T3 (38) — main: func_8001BC6C (69 ins) BANKED byte-identical 143dbb89 via gate_main — permuter_ils found the two levers, the coordinator dropped its wrong-width mutation (R63) and re-spelled them well-defined Drew T 2026-09-05 11:35:41 -06:00
  • 2fdcc4f120 docs(phase-32): T3 S83 CHECKPOINT — 28 banks (census 44 -> 16 stubs / 3,827 ins), fleet R22 218/218 twice, the 17-row Haiku queue 17/17 MATCH; cookbook §500-F; 🛑 block refreshed Drew T 2026-09-05 11:11:03 -06:00
  • fc922a4ccb feat(phase-32): T3 (34) — md_MAIN_009: func_800CD520 (22 ins) BANKED byte-identical d270f695 — §304 self-defining rodata (D_800CD364, 8 words incl. the island's trailing junk) via the ov_SC03_011:0x8017dbd8 twin; the agent's (void*) prototype for func_8001AD38 re-spelled to the TU's (const char*) after the sibling banks landed (§376) Drew T 2026-09-05 11:02:03 -06:00
  • f461715d05 feat(phase-32): T3 (37) — md_SC03_056: func_801CBB80 (23 ins) BANKED byte-identical bc768a6b via the ov_SC06_025:0x8017ded8 twin (d=5) — the module is 100% C Drew T 2026-09-05 11:01:08 -06:00
  • 582bcab70e feat(phase-32): T3 (35)+(36) — md_SC03_054: func_801F0048 (20 ins) + func_801F00C8 (15 ins) BANKED byte-identical 06bd73df — the module is 100% C Drew T 2026-09-05 11:01:07 -06:00
  • f20e42259f feat(phase-32): T3 (27)–(33) — md_MAIN_009: seven Haiku rows BANKED byte-identical d270f695 (func_800CD3B8 21, func_800CD408 23, func_800CD464 25, func_800CD4C8 22, func_800CD558 24, func_800CD5B8 23, func_800CD614 22) Drew T 2026-09-05 11:00:44 -06:00
  • 5160775a14 feat(phase-32): T3 (26) — md_MAIN_007: func_800CF220 (19 ins) BANKED byte-identical 2ff702b6 — Haiku via the ov_SC03_001:0x80182860 twin (d=4) Drew T 2026-09-05 11:00:43 -06:00
  • 5aab66b2de feat(phase-32): T3 (21)–(25) — md_MAIN_007: five Haiku rows BANKED byte-identical 2ff702b6 (func_800CF0F4 21, func_800CF1CC 21, func_800CF268 21, func_800CF33C 21, func_800CEFF8 13) Drew T 2026-09-05 11:00:17 -06:00
  • 8ca9d53cad feat(phase-32): T3 (15)–(20) — md_MAIN_007: six Haiku MATCH rows BANKED byte-identical 2ff702b6 (func_800CEEFC 25, func_800CEF94 25, func_800CF068 20, func_800CF148 33, func_800CF2BC 32, func_800CF3B0 22) Drew T 2026-09-05 10:40:25 -06:00
  • 9e84ce2cd8 chore(phase-32): T3 plumbing — md_MAIN_007: func_800CF3B0 decl → no-proto extern s32 func_800CF3B0(); (byte-neutral, sha 2ff702b6 unchanged) Drew T 2026-09-05 10:37:22 -06:00
  • aa773c8fb0 feat(phase-32): T3 (14) — md_SC03_054: func_801EF6D8 (604 ins + 7 jump tables) BANKED byte-identical 06bd73df — no carve: §303 island pads derived at build Drew T 2026-09-05 10:36:54 -06:00
  • e2aee8e274 feat(phase-32): T3 (12)+(13) — main: func_80015B6C (120 ins) + func_8002FDE8 (73 ins) BANKED byte-identical 143dbb89 via gate_main Drew T 2026-09-05 10:34:35 -06:00
  • 9887e69872 feat(phase-32): T3 (11) — md_SC03_053: func_801EF7E4 (72 ins) BANKED byte-identical c0848f30 — the module is 100% C Drew T 2026-09-05 10:33:03 -06:00
  • d3f05998d5 feat(phase-32): T3 (10) — md_SC03_053: func_801EF734 (44 ins) BANKED byte-identical c0848f30 — §224 pinned pointer + three literal uniform arms for cross_jump Drew T 2026-09-05 10:32:35 -06:00
  • 11f79c1b0a docs(phase-32): T3 checkpoint — the 31 agents' FULL final reports saved to .run/P32/t3/reports/ (tools/agent_reports.py, NEW) + tools/transcript_dump.py (NEW, the dead-session reader); SETUP rows, playbook addendum-2, 🛑 block §3/§6 point at them Drew T 2026-09-05 10:11:43 -06:00
  • 42ff6116ea docs(phase-32): session-start protocol → phase-ends/DIGEST.md + last 3 PhaseEnds (~100k tokens), checkpoint replayed VERBATIM (R64 candidate); the T3 checkpoint refreshed to the verbatim-replay standard Drew T 2026-09-05 10:02:04 -06:00
  • 9480c979a0 docs(phase-32): T3 CHECKPOINT — the launching coordinator overflowed after bank 9; 22 agent verdicts recovered (20 MATCH / 9 NEAR / 2 FAIL of 31), 11 swept Opus deliverables restored + re-verified, cookbook §500 harvest, playbook §S80 addendum-2, agent_drafts_restore.py Drew T 2026-09-05 09:36:43 -06:00
  • 0240fa5f62 feat(phase-32): T3 (9) — main: func_80038698 (74 ins) BANKED byte-identical 143dbb89 via gate_main — the "sched1 DAG-priority + self-coalesce wall" was VARIABLE SCOPE Drew T 2026-09-05 01:07:13 -06:00
  • 0276aaec68 feat(phase-32): T3 (8) — main: func_80023BF0 (281 ins) BANKED byte-identical 143dbb89 via gate_main — the OT link is libgpu's P_TAG 24-bit BITFIELD store (§364), not a hand-written mask/or Drew T 2026-09-05 01:06:17 -06:00
  • ec8a305d75 feat(phase-32): T3 (7) — md_MAIN_007: func_800CEE2C (52 ins) BANKED byte-identical 2ff702b6 (Sonnet drafter; an early u8 *base local keeps &D_800AF630 in $s0 across two calls, the neighbour global read by raw offset with no relocation; one captured pointer local for a shared address) Drew T 2026-09-05 01:05:31 -06:00
  • c6c1352d20 feat(phase-32): T3 (6) — md_SC03_053: func_801EF624 (35 ins) BANKED byte-identical c0848f30 (Haiku drafter; a $16 register pin holds the state pointer across both arms) Drew T 2026-09-05 01:05:04 -06:00
  • d50576f729 feat(phase-32): T3 (5) — md_SC03_053: func_801EF6B0 (33) + func_801EF95C (20) BANKED byte-identical c0848f30 (Haiku drafters: a switch on the u16 state; a twin remap + one guarding beqz) Drew T 2026-09-05 01:04:27 -06:00
  • 3995237eb0 feat(phase-32): T3 (4) — main: func_80039B20 (79 ins) BANKED byte-identical 143dbb89 via gate_main — the S80 permuter plateau at 7 was an ALIAS FLAG, not a scheduler tie Drew T 2026-09-05 01:03:40 -06:00
  • 5ff842dfd3 feat(phase-32): T3 (3) — resident: func_800D06E8 (344 ins) BANKED byte-identical 8e17e02f — THE RESIDENT IS 100% C (145/145) Drew T 2026-09-05 01:03:07 -06:00
  • ceeb1a1eb2 feat(phase-32): T3 (2) — md_SC03_053: func_801EF49C (49 ins) BANKED byte-identical c0848f30 (Haiku drafter; explicit goto labels reproduce the beqz/beq layout) Drew T 2026-09-05 01:01:13 -06:00
  • 4ecedb40ba feat(phase-32): T3 (1) — md_SC03_054: func_801EF558 (96 ins) BANKED byte-identical 06bd73df (Sonnet drafter, first draft; range-dispatch state setter, one reused local keeps the literal 6 in $s2 across the jals) Drew T 2026-09-05 01:01:13 -06:00
  • 3f8f0a5fe3 docs(phase-32): T3 LAUNCHED — 47 targets, 20 drafters running (16 Opus / 3 Sonnet / 1 Haiku), 27 Haiku queued behind the 20-concurrent cap; recovery route logged Drew T 2026-09-05 00:59:06 -06:00
  • 7ce1cd1801 docs(phase-32): T2c CLOSE — 20 free banks, R22 218/218 rc 0, census 54 stubs / 6,376 ins; disc-completeness P32 section (parked ledger EMPTY, audit-disc UNCLAIMED 0 of 220); regenerated fleet/disc/dup digests; T2d not needed Drew T 2026-09-05 00:49:07 -06:00
  • a2d28df189 feat(phase-32): T2c (4) — 7 more free banks in md_SC03_053 (byte-identical c0848f30); the five new modules are now 20/56 banked for 0 drafting tokens Drew T 2026-09-05 00:12:48 -06:00
  • 9ffd0b7ee3 feat(phase-32): T2c (3) — 5 more free banks (md_MAIN_007 ×2, md_SC03_054 ×2, md_SC03_056 ×1), byte-identical Drew T 2026-09-05 00:12:09 -06:00
  • 89d177aa37 feat(phase-32): T2c (2) — 8 free banks in the five new modules (twin remaps, 0 drafting tokens), every module byte-identical Drew T 2026-09-05 00:10:33 -06:00
  • 8b2bbff831 fix(phase-32): T2c (1) — split_indicator's population is derived from the yamls (was a stale stored 213-name list; tools-health said "213 OK of 213" over 218 binaries); R32 denominator assertion Drew T 2026-09-05 00:08:44 -06:00
  • 6853c0c341 docs(phase-32): cookbook §499 (static base evidence for a never-onboarded payload; the first build is a NULL oracle for fine base errors) + decision-log P32 S81 (R31: six instrument defects, one shape) Drew T 2026-09-04 23:55:26 -06:00
  • 6501c74652 docs(phase-32): T2b CLOSE — all five parked payloads onboarded (fleet 213 -> 218, +56 stubs); CURRENT_PHASE log + checkpoint Drew T 2026-09-04 23:53:35 -06:00
  • 1e843c607a feat(phase-32): T2b (4) — SC03/56 ONBOARDED as md_SC03_056 @0x801CBB50 (ov_SC03_002's DESTPTR), byte-identical bc768a6b; ALL FIVE parked payloads are now binaries (fleet 213 -> 218); evidence tool v2 Drew T 2026-09-04 23:52:58 -06:00
  • b52d67be0b feat(phase-32): T2b (3) — SC03/53 + SC03/54 ONBOARDED as md_SC03_053 / md_SC03_054 @0x801EF468 (the script slot), byte-identical c0848f30 / 06bd73df Drew T 2026-09-04 23:51:45 -06:00
  • 10aaf5c296 feat(phase-32): T2b (2) — MAIN/9 ONBOARDED as md_MAIN_009 @0x800CD348 (TEXT_LO 0x3C), byte-identical d270f695; the OPDEMO1 module leaves the parked ledger Drew T 2026-09-04 23:51:08 -06:00
  • 1a696a851a feat(phase-32): T2b (1) — MAIN/7 ONBOARDED as md_MAIN_007 @0x800CEDF8 (TEXT_LO 0x34), byte-identical 2ff702b6; the OPDEMO0 module leaves the parked ledger Drew T 2026-09-04 23:49:05 -06:00
  • f698cf7630 docs(phase-32): T1b/T1c/T2a CLOSE — 3 banks, census 21 -> 18, R22 213/213 rc 0 after the T1b+T1c batch; CURRENT_PHASE log + checkpoint Drew T 2026-09-04 23:43:34 -06:00
  • 848c7c50ab feat(phase-32): T2a — tools/payload_base_evidence.py (controls-gated static base evidence) + memory-map §S45 p7: the parked five get candidate bases Drew T 2026-09-04 23:42:53 -06:00
  • 059266afca feat(phase-32): T1c — md_MAIN_034: func_800CB00C (123 ins) BANKED byte-identical 46153c06 from the stored S72 body; its S68 "compiler wall" pin dropped Drew T 2026-09-04 23:38:03 -06:00
  • c7f5b498ac feat(phase-32): T1b (3) — ov_SC02_017: func_80186C64 (209 ins) BANKED byte-identical c0253499 — the d=2 RELOC-ONLY twin of ov_SC02_016:func_801810C8 via family_remap + four TU spellings Drew T 2026-09-04 23:36:52 -06:00
  • 599dba3ba0 feat(phase-32): T1b (2) — ov_SC02_017: func_80186C64 isolated into its own code subseg [0x5eb0c, c, ov_SC02_017_jr_80186C64] (byte-identical c0253499) Drew T 2026-09-04 23:35:19 -06:00
  • 287ea155c9 docs(phase-32): T1a CLOSE — resident func_800D128C banked, R22 213/213 rc 0; CURRENT_PHASE log + checkpoint (T1b/T1c stored MATCH bodies found via the journals, R38) Drew T 2026-09-04 23:33:34 -06:00
  • c513e1fbbd feat(phase-32): T1a (2) — resident: func_800D128C (243 ins) BANKED byte-identical 8e17e02f via the raw splice + a 5-piece carve; three instrument fixes (§498) Drew T 2026-09-04 23:28:59 -06:00
  • c52190ca86 fix(phase-32): T1b (1) — jr_isolate_all keys a bodiless typedef struct Tag Alias; by the ALIAS (§497); ov_SC02_017 dry-run REFUSED -> CLEAN, no source rename Drew T 2026-09-04 23:24:02 -06:00
  • 380ccdc843 feat(phase-32): T1a (1) — resident code subseg split (3 regions, byte-identical 8e17e02f) + jr_isolate_all include-derived provided types (§496) Drew T 2026-09-04 23:20:04 -06:00
  • a16163ebb6 chore(phase-32): T0 OPEN — Phase 32 (the last 21 + the parked 5) plan approved at gate 1; CURRENT_PHASE.md + baseline reads all green Drew T 2026-09-04 23:12:57 -06:00
  • 77df1092e2 docs(phase-31): post-close bank — accelerators.md S80 (five: loop tools must prove they iterated; the byte gate is a null oracle for 'is this C?'; a pointer-only callee's extern is unconstrained; banked-but-not-merged printed as success; Agent-tool drafters outlive the session) + wave-playbook S80 (parallel_gate's exit-2 guard and recovery route in the gate section; the one-agent-per-function shape + agent_verdicts addendum) Drew T 2026-09-04 22:36:55 -06:00
  • 27180c705d chore(phase-31): CLOSE — the frontier atlas & wide-tolerance campaign; fleet 95.3% -> 100.0% instr (213/213), main 1,041 -> 12 open, the last 21 ledgered (v1.30.0) Drew T 2026-09-04 22:26:47 -06:00
  • 3c292ef097 docs(phase-31): S80 — docs/frontier-p32.md, the Phase-32 jumping-off document: the last 21 functions by blocker (each with its instrument's verdict, mechanism citation, best-draft path and route — the two NON-CONTIGUOUS carves dry-run/refusal detail, the RELOC-ONLY twin, the leading-island case, the 8 pinned walls, the 3 near plateaus, the 7 far drafts), the five parked disc payloads, the routes/instruments a fresh session must know, and the proposed Phase-32 shape with its kill gate Drew T 2026-09-04 22:20:20 -06:00
  • 452975e852 docs(phase-31): S80 #10 CLOSE — the verbatim end-state: manifest 6 → 5 rows (the five PERMANENT rows RATIFIED in _README; the GAME-C row decompiled), cookbook §495 (two def-side declaration walls, the S79 assembly "bank" P9 correction, the gate that dropped a bank on exit 0), decision-log S80 addendum (R31), SETUP rows; tools/parallel_gate.py: banked-but-not-merged now exits 2 with the worker's raw git status kept + per-run .run/pgate_runs/<ts>.json; CURRENT_PHASE #10 bullet + the S80 #10 CLOSE checkpoint (R22 213/213, tools-health OK, census 21 stubs / 4,554 ins, NEXT #11 = PhaseEnd, gate 2); regenerated digests Drew T 2026-09-04 22:07:35 -06:00
  • 082ddc8590 feat(phase-31): S80 #10 (2) — ov_SC03_107: func_8017D878 (45 ins, the manifest's one GAME-C verbatim, "37 stored drafts") DECOMPILED: the stored best draft was byte-correct all along — it needed only its real return type (s32: the address of func_80172710 or 0) once the TU's extern void func_8017D878(void) was corrected (commit:3893) — a def-side declaration wall, not codegen; match_one 45/45, rtu_match MATCH, worktree gate banked (its merge step adopted 0 files — spliced in-tree instead) + in-tree byte-identical 87d02b57. ov_SC03_107 is now 100% C; the game's verbatim set is the five PERMANENT rows Drew T 2026-09-04 21:49:34 -06:00
  • b7f73e7187 feat(phase-31): S80 #10 (1) — md_MAIN_020: func_800CB17C (30 ins) DECOMPILED — the S79 #7 "raw splice" had banked its assembly as a verbatim __asm__ body (P9 correction: that was not a decompile, verbatim_check --strict caught it as a new verbatim); the C is seven straight calls (s0 = a0; func_8012A304(*(s16*)(s0+0xA), 0x80); func_801477E8(s0, 0xFFE80000); func_80154274(s0, D_800CB1F4); func_80147324(0x47A); func_8014C010(s0, 2); func_801483E8(s0); func_80019064(D_80062BD0)) as an s32 function with no return (the target sets no $v0); the TU's block-scope struct S decl needed a file-scope tag (commit:3894); match_one 30/30, rtu_match MATCH, worktree gate + in-tree byte-identical 0990e041 Drew T 2026-09-04 21:49:33 -06:00
  • ec15509e68 chore(phase-31): S80 #10 — md_MAIN_020: a file-scope struct S; tag so the TU's block-scope extern s32 func_800CB17C(struct S *) no longer declares a block-local tag that no file-scope definition can match (conflicting types); byte-identical 0990e041 Drew T 2026-09-04 21:46:55 -06:00
  • dd45617982 chore(phase-31): S80 #10 — verbatim end-state plumbing, byte-neutral: (1) ov_SC03_107:func_8017D878's file-scope __asm__ block (the manifest's one GAME-C row) converted back to an INCLUDE_ASM stub with tools/verbatim_to_stub.py --gate (87d02b57 byte-identical) and the TU's extern void func_8017D878(void) corrected to the real s32 (s32) signature (only its address is taken; ov_SC03_107 rebuilt byte-identical) — the void decl was the def-side wall that refused 37 drafts; (2) md_MAIN_020:func_800CB17C — S79 #7's "raw splice" had committed the function's ASSEMBLY as a verbatim __asm__ body (every stored 'draft' was the asm; verbatim_check --strict flagged it as a new verbatim) → converted back to a stub (0990e041 byte-identical): that bank was not a decompile (P9), the function is open again; (3) manifest _README: the five PERMANENT-VERBATIM rows RATIFIED Drew T 2026-09-04 21:45:31 -06:00
  • 846c6d5891 docs(phase-31): S80 #9 CLOSE checkpoint — task #9 DONE (ten banks; open stubs 31 → 21; main REAL 777 / stubs 12 / game-code 95.1%; fleet instr 100.0%, 213/213 R22, tools-health OK); the frontier census regenerated (21 = 4 B-CARVE + 10 D-NEAR + 7 F-FAR, 4,554 ins); NEXT #10 with its design brief (main() is already C, the -O0 cluster is spent — #10 = ratify the 5 PERMANENT verbatims + decompile ov_SC03_107:func_8017D878); regenerated digests Drew T 2026-09-04 21:41:17 -06:00
  • 09e0b27811 docs(phase-31): S80 #9c — the S79 drafting task's NEAR/WALL ledger: 15 rows logged to the backlog (each with the residual's gcc mechanism), the ≤3 residuals pinned as WALL candidates in config/wave_exclude.txt after an 8-seed permuter_ils sweep on the now-permutable pinned seeds (no score-0; func_80039DEC 9→2 = the K&R raw-preserve register, func_80023BF0 18→11 ADDRESSING, func_8017DF28's "1" was a divergent store rewrite — closeness stays 2, R14); cookbook §494 v2 (ten banks, the Opus verdicts' idioms: P_TAG bitfield store, inverted arms, sibling-reading, the 518-ins spelling laws, gdb-on-cc1 allocno arithmetic, K&R s16 params, true_dependence, field-boundary fences, extendhisi2 orphans, [][1] decls) + the final ledger; .run/S79w allowlisted (27 drafts + 3 permuter waypoints + verdict ledger + briefs, 264 KB — R20) Drew T 2026-09-04 21:21:05 -06:00
  • 7fbdb8fd63 fix(phase-31): S80 #9c — the permuter could not permute a PINNED seed, and it was our instrument: hide_asm carried only the __asm__ spelling (3 S79 seeds use asm("$7")), permuter_ils warm-restarted from the DECODED waypoint (raw pins back in base.c → cycles 2..N were silent parser refusals reported "(unchanged)"), and defines_fn refused K&R-style definitions (436 stored backlog drafts kept out of the lane for four phases). Fixed + R39-controlled over 5,311 drafts (the bare word asm in INCLUDE_ASM path strings was a caught false positive): re-hide every waypoint, assert the definition survived, abort exit-2 on a refusal (R61a), flushed logs (R55). Every S79 pinned seed now iterates; ov_SC06_022:func_8017DF28 (pinned WALL, closeness 2) reached 1 in its first cycle. cookbook §493 S80 correction + §494 v1 (S79 idioms); SETUP rows (p16_permute/permuter_ils, agent_verdicts.py) Drew T 2026-09-04 20:59:35 -06:00
  • 402b7520c4 feat(phase-31): S79 #9 (10) — ov_SC02_027: func_80180B3C (297 ins, F-FAR) banked from an Opus agent's draft: recovered the S76 closeness-23 body and broke the 5-attempt plateau with four dials — tpg |= (y & 0x200) << 2; as its own accumulator statement, a register u32 c40 __asm__("$2") pin on the (w&0x40)>>6 term, the if (c) v -= 0x100 split into a second variable so sched1's birthing boost stops sinking the mask, and the inverted-arm if (!c) vv = v; else vv = v - 0x100; that keeps the &0xFFFF/-0x100 alive for the u8 store (the same arms un-inverted delete the andi); statement order INERT (792 permutations); real-TU MATCH after dropping the TU's duplicate Blk32_80180908 typedef; worktree gate + in-tree byte-identical Drew T 2026-09-04 20:50:13 -06:00
  • d04fe31882 feat(phase-31): S79 #9 (9) — main: func_8001EFE0 (468 ins, the largest G-UNKNOWN main body) banked from an Opus agent's draft: recovered the prior closeness-14 body from .run/match and closed all three [permuter]-filed clusters by reading the matched same-TU siblings func_8001DA34/func_8001EA14 — the addPrim tail's ot split so the sll lands between the two pinned luis, one zero-byte fence between the tpage sh and the q[7] RMW, a §419 density asm on (vh,vv2,vw) for the two-SVECTOR fill, and the shift split from its mask; rtu_match --tu src/800.c MATCH, gate_main --apply clean rebuild BYTE-IDENTICAL 143dbb89 Drew T 2026-09-04 20:47:51 -06:00
  • 618e0907d4 feat(phase-31): S79 #9 (8) — ov_SC01_001: func_80181E04 (269 ins, the GAME-GTE 'uncertain' body) banked as compiler C by an Opus agent: sixteen D_801EDA?? globals as one 0x50-stride record array (§246-2/§200), a counted i<0x100 loop, the gte_stsz3 branch shape, and the OT insert as PsyQ's P_TAG addr:24 bitfield store (store_bit_field masks the value first); real-TU MATCH, worktree gate + in-tree byte-identical Drew T 2026-09-04 19:57:19 -06:00
  • 2709321082 docs(phase-31): S79 HANDOFF checkpoint mid-task #9 — 7 banks this task (open stubs 51 -> 25 this session), plateaus with residuals named, 11 drafting agents still running; tools/agent_verdicts.py extracts their final JSON verdicts from the subagent transcripts for the fresh session to aggregate (procedure + paths in the 🛑 block) Drew T 2026-09-04 19:50:01 -06:00
  • 96c0fc02a7 feat(phase-31): S79 #9 (7) — ov_SC05_010: func_8017FFA8 (88 ins, 6-way jtbl switch) banked from a Sonnet agent's fresh draft (loop index s32 not s16 — §241 fused sign-extend; the D_801922B8 lookup into its own temp before the found/zero stores); tail jtbl carve at gate time; worktree gate + in-tree byte-identical Drew T 2026-09-04 19:45:16 -06:00
  • cdf50f96d7 feat(phase-31): S79 #9 (6) — ov_SC06_010: func_801809E4 (33 ins) banked by the Sonnet escalation (the Haiku plateau's missing 2 ins was an early return that should fall through into the shared mask/store tail; the -0x10 no-save frame is two s16 locals, §186b); real-TU MATCH, worktree gate + in-tree byte-identical Drew T 2026-09-04 19:43:50 -06:00
  • 1961a7bfb1 feat(phase-31): S79 #9 (5) — main: func_8002AC98 (114 ins) banked: the generic arm's val = r + b routed through a fresh s32 temp ({ s32 xt = r + b; val = xt; }, with b widened to s32) so expand_binop's target==op1 swap does not fire — the same lever as func_80015608; gate_main BYTE-IDENTICAL 143dbb89 Drew T 2026-09-04 19:05:29 -06:00