mirror of
https://github.com/Druthulu/BFM-decomp
synced 2026-09-28 14:59:48 -04:00
Drew T
93d2851ab2
- Autopsy 1,349 rows: 1,178 near / 63 nobuild / 108 match_one-MATCH (T1 fuel, 3x the S16 sample); 0 classifier errors; second-oracle agree. - THE -O0 PIN: 18 fam / 2,192 open members = 2,131 o0b-route-eligible + 61 SC07-only; 15/18 exemplar-matched. The 'WALLED ~1,287' framing dissolves — T2 generalizes rollout_801457a4_o0.py (proven 130/130); jr+o0 composites (0x8013c414, 0x8013c0f8) probe x1 first. - Zero-crack: 114 fam / 3,166 members / ~204k w-ins (28 substantial). Concentration top-20 24.0% / top-100 44.8%. Worklist fresh: 101 live / 261,789 gain-ins (stale 160/583k gone). - T0 COMPLETE (a-f) — report point #1 = docs/frontier-p30.md
75 lines
4.3 KiB
Markdown
75 lines
4.3 KiB
Markdown
# Autopsy — the permuter's failures, classified
|
|
|
|
> Generated by `tools/autopsy.py report` from `.run/autopsy/residuals.jsonl` (built by `collect`, which recompiles every open backlog draft through `tools/match_one.py` and classifies the FULL residual with `tools/residual_class.py`).
|
|
|
|
> A verdict here is a **candidate**, never a bank: the whole-binary byte-gate remains the sole arbiter (G3/P9).
|
|
|
|
**Corpus:** 1349 record(s) · 1286 classified · status {'near': 1178, 'nobuild': 63, 'match': 108}
|
|
|
|
|
|
## Buckets — what tool each failure actually wants
|
|
|
|
| bucket | fns | reach-weighted | meaning |
|
|
|--------|----:|---------------:|---------|
|
|
| redraft | 669 | 1068 | the stored draft is NOT this function — un-attempted work misfiled as a near-miss; route to a fresh crack, not to a wall |
|
|
| structural | 483 | 3551 | local mutation CANNOT introduce it — wants a C-level idiom, not CPU |
|
|
| integration | 108 | 1446 | byte-correct standalone; blocked on plumbing (§58/§59 reconcile ladder) |
|
|
| permuter | 24 | 157 | local mutation CAN reach it — permuter fuel (a plateau here is a **missing-transform** candidate) |
|
|
| unknown | 2 | 2 | the residue the LLM tier is for |
|
|
|
|
## Classes
|
|
|
|
| class | fns | reach-weighted | bucket | permuter profile |
|
|
|-------|----:|---------------:|--------|------------------|
|
|
| SIZE-MISMATCH | 669 | 1068 | redraft | — |
|
|
| LENGTH-DRIFT | 307 | 2843 | structural | — |
|
|
| MATCH | 108 | 1446 | integration | — |
|
|
| OPCODE-MIXED | 70 | 336 | structural | — |
|
|
| WIDTH | 39 | 305 | structural | — |
|
|
| IMM-OFFSET | 36 | 36 | structural | — |
|
|
| BRANCH-POLARITY | 18 | 18 | structural | — |
|
|
| ADDRESSING | 15 | 15 | structural | cse |
|
|
| DELAY-SLOT | 12 | 145 | permuter | schedule |
|
|
| REGALLOC-PERM | 6 | 6 | permuter | regalloc |
|
|
| IMM-VALUE | 4 | 4 | permuter | cse |
|
|
| MIXED | 2 | 2 | unknown | — |
|
|
|
|
## Top clusters (the batch-autopsy work items)
|
|
|
|
One representative per signature diagnoses the whole cluster — that is what makes the LLM pass a *batch* autopsy rather than N per-function reads.
|
|
|
|
| # | signature | fns | reach-weighted | example |
|
|
|--:|-----------|----:|---------------:|---------|
|
|
| 1 | `SIZE-MISMATCH/short` | 594 | 993 | func_8017FF90 (close=4, 2 ins) |
|
|
| 2 | `LENGTH-DRIFT/-1?` | 108 | 777 | func_8018312C (close=5, 7 ins) |
|
|
| 3 | `MATCH` | 108 | 1446 | func_80168B70 (close=0, 27 ins) |
|
|
| 4 | `SIZE-MISMATCH/long` | 75 | 75 | func_8017E06C (close=6, 9 ins) |
|
|
| 5 | `OPCODE-MIXED` | 61 | 327 | func_80161C24 (close=1, 29 ins) |
|
|
| 6 | `LENGTH-DRIFT/-2?` | 57 | 856 | func_80190480 (close=3, 4 ins) |
|
|
| 7 | `LENGTH-DRIFT/1?` | 55 | 454 | func_80187074 (close=4, 7 ins) |
|
|
| 8 | `LENGTH-DRIFT/2?` | 42 | 308 | func_80181B8C (close=5, 11 ins) |
|
|
| 9 | `LENGTH-DRIFT/-7` | 32 | 32 | func_80184590 (close=9, 2 ins) |
|
|
| 10 | `IMM-OFFSET/1` | 17 | 17 | func_801827C4 (close=1, 13 ins) |
|
|
| 11 | `WIDTH/lhu!=lh` | 9 | 9 | func_80142A10 (close=1, 28 ins) |
|
|
| 12 | `DELAY-SLOT/1` | 9 | 142 | func_8016BA68 (close=1, 29 ins) |
|
|
| 13 | `BRANCH-POLARITY/beqz!=bnez` | 8 | 8 | func_8017FDB0 (close=1, 12 ins) |
|
|
| 14 | `BRANCH-POLARITY/bnez!=beqz` | 7 | 7 | func_80185E38 (close=1, 12 ins) |
|
|
| 15 | `OPCODE-MIXED/addressing,width` | 5 | 5 | func_8017F114 (close=5, 75 ins) |
|
|
| 16 | `ADDRESSING/lui!=addiu` | 5 | 5 | func_8017D468 (close=3, 13 ins) |
|
|
| 17 | `WIDTH/sh!=lw` | 5 | 5 | func_8018AFDC (close=3, 8 ins) |
|
|
| 18 | `WIDTH/lw!=lhu` | 4 | 270 | func_8015FBE0 (close=1, 58 ins) |
|
|
| 19 | `IMM-VALUE` | 4 | 4 | func_801879B0 (close=2, 10 ins) |
|
|
| 20 | `WIDTH/sw!=sh` | 4 | 4 | func_80185000 (close=2, 12 ins) |
|
|
| 21 | `REGALLOC-PERM/$a0>$a1` | 3 | 3 | func_801863D4 (close=1, 8 ins) |
|
|
| 22 | `IMM-OFFSET/-3` | 3 | 3 | func_80182A8C (close=1, 13 ins) |
|
|
| 23 | `WIDTH/sw!=addiu` | 2 | 2 | func_8017F474 (close=8, 14 ins) |
|
|
| 24 | `LENGTH-DRIFT/-4?` | 2 | 135 | func_80161A90 (close=26, 30 ins) |
|
|
| 25 | `IMM-OFFSET/-48` | 2 | 2 | func_801807D8 (close=1, 4 ins) |
|
|
|
|
## Second oracle — deterministic verdict vs the recorded label
|
|
|
|
The recorded `klass` is what a wave agent typed months ago; the verdict is derived from today's bytes. Neither is authoritative — **the disagreement is the finding**, and it is exactly the `mis-classified` bucket (a wrongly-labelled residual sends the directed search at the wrong levers, so it plateaus and then reads as a wall).
|
|
|
|
- labelled records comparable: **1** · agree **1** · disagree **0**
|
|
|