mirror/BFM-decomp
1
0
Fork 0
mirror of https://github.com/Druthulu/BFM-decomp synced 2026-09-30 15:36:55 -04:00
Code Issues Packages Projects Releases Wiki Activity

feat(phase-29 Task-13B close): plateau autopsy = ZERO missing-transforms; narrow the admission instead

Browse Source 
The hindsight-study §7 taxonomy predicts plateaus decompose into missing-transform (the
"highest-value bucket and the whole point"), seed-structural, and genuine-wall. Run against
real plateaus this class produced NO missing-transforms, and the answer needed no LLM.

MEASURED: `length` probe, 20 targets, 1 win. tail 1/6; partial 0/12.
AUTOPSY (read directly from the bytes, 3 partial plateaus):
- func_8017F0C0 / func_801806C8: target has `sltiu $v0,$v0,1` = gcc's codegen for `!x`/`x==0`;
  the drafts wrote `(u32)(D_x ^ 1)` which emits `xori`. No local mutation crosses that.
- func_8017FF90: draft stores to arg0+8, target stores to a GLOBAL. Different function.
=> these are WRONG DRAFTS wearing a small closeness, i.e. seed-structural, not a mutation gap.

THE FIX IS THE OPPOSITE OF "ADD TRANSFORMS" — a tighter ADMISSION rule:
- _drift_route: permuter only when |d|<=2 AND explains=="tail" (the shape that measurably
  converts). length pool 339 -> 34; permuter bucket 389 -> 84.
- SIZE-MISMATCH: added a PROPORTIONAL test (|d| >= 0.5*nt). max(2,0.15*nt) is far too
  permissive on a tiny target — a 2-ins draft vs a 4-ins target read as a near-miss.
permuter_weights needs NO extension for this class.

Transferable (cookbook §60b): raising a search-closer's yield is at least as often about
refusing it unreachable work as widening its mutation set. Same knife as Task-13A's
targeting fix, one cut finer. Drafter idiom recorded: `sltiu rd,rs,1` => `!x`, never `x^1`.

17 unit tests green; corpus re-collected (1654 rows, closeness cross-check clean).
This commit is contained in:
Drew T
2026-07-21 17:45:49 -06:00
parent 83f20f4271
commit 835c45905d
5 changed files with 1436 additions and 1376 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/backlog.md
+1219 -1220
View File
File diff suppressed because it is too large Load Diff
docs/matching-cookbook.md
+35
View File
@@ -4479,3 +4479,38 @@ diverged.
path, because nothing exercises it. Before trusting an unattended fix-and-run, budget for the first success
to fail — and check the tree state, not the exit code (here the durable Task-12 winner save in
`.run/permuter-winners/` is what made the crash a non-event).
### §60b — The plateau autopsy's verdict: a `partial` drift is a WRONG DRAFT, not a missing transform (Phase 29 Task-13B close, 2026-07-21)
hindsight-study §7 predicts that a search-closer's plateaus decompose into *missing-transform* (extend the
mutation set — "the highest-value bucket and the whole point"), *seed-structural*, and *genuine-wall*. Run
against real plateaus, **this class produced ZERO missing-transforms.** The autopsy is worth recording
because the answer was legible in the bytes and needed no LLM at all.
**The measurement.** A 20-target probe of the `length` profile: `tail` drifts (a single shift point explains
the whole tail) converted **1/6**; `partial` drifts (length differs AND other positions differ) converted
**0/12**. Reading three `partial` plateaus directly:
* `func_8017F0C0`, `func_801806C8` — target contains `sltiu $v0,$v0,0x1`. **That is gcc's codegen for
`!x` / `x == 0`.** The drafts wrote `(u32)(D_x ^ 1)`, which emits `xori`. No local mutation rewrites
`xori` into `sltiu`: it is a different operation, chosen by the front end from a different C expression.
* `func_8017FF90` — the draft stores to `arg0 + 8`; the target stores to a **global**
(`lui $at,%hi(D_…)` / `sw $zero,%lo(D_…)($at)`). Not the same function at all.
**Two permanent fixes, both NARROWING what the offline tool is allowed to attempt:**
1. `_drift_route` now admits a drift to the permuter only when `|Δ|<=2` **AND** `explains == "tail"`.
Length-profile pool 339 → **34**; permuter bucket 389 → **84**. A `partial` drift is seed-structural by
construction — the count is wrong *and* other positions are wrong, which is not one local edit.
2. `SIZE-MISMATCH` gained a PROPORTIONAL test (`|Δ| >= 0.5*nt`) alongside the absolute one. `max(2, 0.15*nt)`
is far too permissive on a tiny target: a 2-instruction draft against a 4-instruction target is `|Δ|=2`
and read as a near-miss when it is a wholesale mismatch.
**The transferable lesson.** The §7 taxonomy tacitly assumes the plateaus are *near*. Ours mostly were not —
they were bad drafts wearing a small `closeness`. So the highest-value autopsy outcome was not a new
transform but a **tighter admission rule**: the way to raise a search-closer's yield is at least as often to
stop feeding it unreachable work as to widen its mutation set. Same knife as Task-13A's targeting fix, one
cut finer.
**Cookbook idiom for drafters (recurring):** `sltiu rd, rs, 1` ⇒ the C is `!x` / `x == 0`, NOT `x ^ 1`.
The XOR form emits `xori` and can never match.
docs/progress.fleet.md
+143 -143
View File
@@ -4,156 +4,156 @@
# cross-binary collapsible-byte leverage: docs/duplicates.cross.md.
# THREE progress metrics (all matter — see the labels):
FLEET fn-count byte-ident: 310961 / 353722 = 87.91% (REAL+LINKED+empties; FUNCTION-count, ×134-inflated — one crack counts per overlay)
FLEET instr-weighted : 10204272 / 13081451 = 78.0% (shipped .text across resident+138 overlays; the decomp.dev-DISPLAY number)
FLEET distinct-code(uniq): 3708963 / 5574674 = 66.5% (64104/87459 unique fns; the DISTINCT-RE number)
FLEET fn-count byte-ident: 311079 / 353722 = 87.94% (REAL+LINKED+empties; FUNCTION-count, ×134-inflated — one crack counts per overlay)
FLEET instr-weighted : 10207672 / 13081451 = 78.0% (shipped .text across resident+138 overlays; the decomp.dev-DISPLAY number)
FLEET distinct-code(uniq): 3708970 / 5574674 = 66.5% (64105/87459 unique fns; the DISTINCT-RE number)
MAIN game-code weighted : 436 / 60201 = 0.7% (Phase-27 T10; SEPARATE — LINKED-excluding Ghidra sig dated 2026-06-14, PROVISIONAL until a fresh/complete main sig; NOT folded into the fleet number)
FLEET REAL substantive : 309106 (of which dedup-shared 234299 via 1847 groups / 234343 instances)
FLEET REAL substantive : 309224 (of which dedup-shared 234437 via 1848 groups / 234481 instances)
FLEET LINKED PsyQ objs : 959
FLEET NON_MATCHING : 7 (0 in any default build — G4)
FLEET INCLUDE_ASM stubs : 42754
FLEET INCLUDE_ASM stubs : 42636
FLEET matchable : 353722
| binary | REAL | shared | LINKED | byte-ident | matchable | byte-ident % |
|---|---:|---:|---:|---:|---:|---:|
| main | 54 | 2 | 959 | 1055 | 2096 | 50.3% |
| resident | 129 | 0 | 0 | 131 | 145 | 90.3% |
| ov_SC01_000 | 2219 | 1705 | 0 | 2219 | 2403 | 92.3% |
| ov_SC01_001 | 2220 | 1705 | 0 | 2222 | 2466 | 90.1% |
| ov_SC01_004 | 2211 | 1696 | 0 | 2212 | 2414 | 91.6% |
| ov_SC01_005 | 2240 | 1718 | 0 | 2240 | 2503 | 89.5% |
| ov_SC01_006 | 2240 | 1718 | 0 | 2240 | 2503 | 89.5% |
| ov_SC01_008 | 2211 | 1696 | 0 | 2213 | 2426 | 91.2% |
| ov_SC01_009 | 2236 | 1697 | 0 | 2237 | 2507 | 89.2% |
| ov_SC01_074 | 2214 | 1697 | 0 | 2216 | 2425 | 91.4% |
| ov_SC01_077 | 2405 | 1665 | 0 | 2407 | 2585 | 93.1% |
| ov_SC01_080 | 2254 | 1700 | 0 | 2254 | 2512 | 89.7% |
| ov_SC01_084 | 2260 | 1700 | 0 | 2265 | 2579 | 87.8% |
| ov_SC02_000 | 2314 | 1735 | 0 | 2314 | 2683 | 86.2% |
| ov_SC02_003 | 2314 | 1735 | 0 | 2314 | 2683 | 86.2% |
| ov_SC02_004 | 2219 | 1700 | 0 | 2219 | 2401 | 92.4% |
| ov_SC02_005 | 2326 | 1696 | 0 | 2336 | 2927 | 79.8% |
| ov_SC02_011 | 2338 | 1703 | 0 | 2349 | 2893 | 81.2% |
| ov_SC02_015 | 2220 | 1702 | 0 | 2220 | 2414 | 92.0% |
| ov_SC02_016 | 2252 | 1702 | 0 | 2255 | 2545 | 88.6% |
| ov_SC02_017 | 2298 | 1702 | 0 | 2306 | 2732 | 84.4% |
| ov_SC02_021 | 2226 | 1702 | 0 | 2226 | 2437 | 91.3% |
| ov_SC02_026 | 2241 | 1700 | 0 | 2247 | 2572 | 87.4% |
| ov_SC02_027 | 2264 | 1700 | 0 | 2273 | 2692 | 84.4% |
| ov_SC02_028 | 2265 | 1700 | 0 | 2275 | 2701 | 84.2% |
| ov_SC02_031 | 2251 | 1701 | 0 | 2256 | 2562 | 88.1% |
| ov_SC02_035 | 2227 | 1700 | 0 | 2230 | 2520 | 88.5% |
| ov_SC02_039 | 2212 | 1700 | 0 | 2212 | 2416 | 91.6% |
| ov_SC02_041 | 2247 | 1700 | 0 | 2250 | 2561 | 87.9% |
| ov_SC03_001 | 2349 | 1701 | 0 | 2366 | 2869 | 82.5% |
| ov_SC03_002 | 2274 | 1705 | 0 | 2289 | 2631 | 87.0% |
| ov_SC03_003 | 2222 | 1700 | 0 | 2223 | 2423 | 91.7% |
| ov_SC03_006 | 2287 | 1706 | 0 | 2296 | 2767 | 83.0% |
| ov_SC03_007 | 2268 | 1700 | 0 | 2272 | 2623 | 86.6% |
| ov_SC03_010 | 2228 | 1700 | 0 | 2228 | 2470 | 90.2% |
| ov_SC03_011 | 2236 | 1700 | 0 | 2242 | 2527 | 88.7% |
| ov_SC03_012 | 2214 | 1700 | 0 | 2215 | 2406 | 92.1% |
| ov_SC03_013 | 2233 | 1700 | 0 | 2233 | 2492 | 89.6% |
| ov_SC03_014 | 2293 | 1725 | 0 | 2293 | 2685 | 85.4% |
| ov_SC03_015 | 2292 | 1725 | 0 | 2292 | 2685 | 85.4% |
| ov_SC03_023 | 2220 | 1700 | 0 | 2221 | 2435 | 91.2% |
| ov_SC03_024 | 2280 | 1705 | 0 | 2287 | 2641 | 86.6% |
| ov_SC03_028 | 2260 | 1700 | 0 | 2264 | 2664 | 85.0% |
| ov_SC03_029 | 2262 | 1702 | 0 | 2272 | 2644 | 85.9% |
| ov_SC03_030 | 2236 | 1705 | 0 | 2238 | 2495 | 89.7% |
| ov_SC03_031 | 2232 | 1700 | 0 | 2235 | 2514 | 88.9% |
| ov_SC03_089 | 2246 | 1700 | 0 | 2253 | 2581 | 87.3% |
| ov_SC03_090 | 2248 | 1700 | 0 | 2256 | 2625 | 85.9% |
| ov_SC03_091 | 2251 | 1700 | 0 | 2259 | 2640 | 85.6% |
| ov_SC03_092 | 2262 | 1700 | 0 | 2274 | 2587 | 87.9% |
| ov_SC03_093 | 2238 | 1700 | 0 | 2242 | 2563 | 87.5% |
| ov_SC03_094 | 2233 | 1700 | 0 | 2239 | 2577 | 86.9% |
| ov_SC03_095 | 2224 | 1700 | 0 | 2227 | 2474 | 90.0% |
| ov_SC03_096 | 2223 | 1700 | 0 | 2226 | 2465 | 90.3% |
| ov_SC03_097 | 2252 | 1700 | 0 | 2259 | 2603 | 86.8% |
| ov_SC03_098 | 2231 | 1700 | 0 | 2234 | 2541 | 87.9% |
| ov_SC03_099 | 2224 | 1700 | 0 | 2227 | 2504 | 88.9% |
| ov_SC03_100 | 2234 | 1700 | 0 | 2238 | 2539 | 88.1% |
| ov_SC03_101 | 2234 | 1700 | 0 | 2238 | 2528 | 88.5% |
| ov_SC03_102 | 2226 | 1700 | 0 | 2229 | 2492 | 89.4% |
| ov_SC03_103 | 2229 | 1702 | 0 | 2232 | 2510 | 88.9% |
| ov_SC03_104 | 2254 | 1700 | 0 | 2261 | 2616 | 86.4% |
| ov_SC03_105 | 2248 | 1700 | 0 | 2255 | 2597 | 86.8% |
| ov_SC03_108 | 2216 | 1700 | 0 | 2216 | 2443 | 90.7% |
| ov_SC03_109 | 2217 | 1700 | 0 | 2219 | 2424 | 91.5% |
| ov_SC03_110 | 2222 | 1700 | 0 | 2222 | 2468 | 90.0% |
| ov_SC03_111 | 2234 | 1700 | 0 | 2237 | 2510 | 89.1% |
| ov_SC03_112 | 2227 | 1702 | 0 | 2229 | 2526 | 88.2% |
| ov_SC03_113 | 2220 | 1702 | 0 | 2223 | 2468 | 90.1% |
| ov_SC03_114 | 2210 | 1700 | 0 | 2212 | 2415 | 91.6% |
| ov_SC03_115 | 2227 | 1700 | 0 | 2229 | 2472 | 90.2% |
| ov_SC03_116 | 2217 | 1700 | 0 | 2220 | 2438 | 91.1% |
| ov_SC03_117 | 2245 | 1700 | 0 | 2251 | 2557 | 88.0% |
| ov_SC03_118 | 2292 | 1724 | 0 | 2293 | 2685 | 85.4% |
| ov_SC03_119 | 2292 | 1724 | 0 | 2293 | 2685 | 85.4% |
| ov_SC03_121 | 2224 | 1700 | 0 | 2227 | 2459 | 90.6% |
| ov_SC03_124 | 2304 | 1696 | 0 | 2324 | 2741 | 84.8% |
| ov_SC03_125 | 2261 | 1700 | 0 | 2273 | 2588 | 87.8% |
| ov_SC03_126 | 2223 | 1702 | 0 | 2223 | 2423 | 91.7% |
| ov_SC04_000 | 2239 | 1704 | 0 | 2248 | 2546 | 88.3% |
| ov_SC04_002 | 2255 | 1700 | 0 | 2259 | 2636 | 85.7% |
| ov_SC04_003 | 2233 | 1700 | 0 | 2237 | 2502 | 89.4% |
| ov_SC04_004 | 2242 | 1700 | 0 | 2244 | 2558 | 87.7% |
| ov_SC04_005 | 2250 | 1700 | 0 | 2256 | 2612 | 86.4% |
| ov_SC04_006 | 2223 | 1700 | 0 | 2225 | 2454 | 90.7% |
| ov_SC04_007 | 2243 | 1700 | 0 | 2247 | 2569 | 87.5% |
| ov_SC04_008 | 2213 | 1700 | 0 | 2213 | 2415 | 91.6% |
| ov_SC04_009 | 2229 | 1700 | 0 | 2232 | 2441 | 91.4% |
| ov_SC04_010 | 2220 | 1700 | 0 | 2221 | 2419 | 91.8% |
| ov_SC04_011 | 2276 | 1700 | 0 | 2282 | 2803 | 81.4% |
| ov_SC04_012 | 2216 | 1700 | 0 | 2217 | 2420 | 91.6% |
| ov_SC04_015 | 2282 | 1701 | 0 | 2293 | 2611 | 87.8% |
| ov_SC04_016 | 2219 | 1700 | 0 | 2221 | 2440 | 91.0% |
| ov_SC04_018 | 2331 | 1735 | 0 | 2331 | 2857 | 81.6% |
| ov_SC04_019 | 2338 | 1735 | 0 | 2338 | 2857 | 81.8% |
| ov_SC04_020 | 2248 | 1700 | 0 | 2260 | 2567 | 88.0% |
| ov_SC04_021 | 2224 | 1702 | 0 | 2224 | 2423 | 91.8% |
| ov_SC05_000 | 2220 | 1704 | 0 | 2223 | 2422 | 91.8% |
| ov_SC05_001 | 2243 | 1700 | 0 | 2248 | 2574 | 87.3% |
| ov_SC05_002 | 2225 | 1700 | 0 | 2228 | 2442 | 91.2% |
| ov_SC05_003 | 2222 | 1700 | 0 | 2223 | 2481 | 89.6% |
| ov_SC05_004 | 2217 | 1700 | 0 | 2219 | 2464 | 90.1% |
| ov_SC05_005 | 2224 | 1700 | 0 | 2225 | 2491 | 89.3% |
| ov_SC05_006 | 2216 | 1700 | 0 | 2216 | 2430 | 91.2% |
| ov_SC05_007 | 2226 | 1700 | 0 | 2231 | 2482 | 89.9% |
| ov_SC05_008 | 2243 | 1700 | 0 | 2245 | 2543 | 88.3% |
| ov_SC05_009 | 2221 | 1700 | 0 | 2225 | 2438 | 91.3% |
| ov_SC05_010 | 2247 | 1700 | 0 | 2251 | 2588 | 87.0% |
| ov_SC05_011 | 2219 | 1700 | 0 | 2220 | 2409 | 92.2% |
| ov_SC05_017 | 2327 | 1697 | 0 | 2338 | 2842 | 82.3% |
| ov_SC05_018 | 2274 | 1700 | 0 | 2287 | 2673 | 85.6% |
| ov_SC05_019 | 2224 | 1702 | 0 | 2224 | 2423 | 91.8% |
| ov_SC06_000 | 2287 | 1709 | 0 | 2290 | 2691 | 85.1% |
| ov_SC06_006 | 2239 | 1700 | 0 | 2240 | 2511 | 89.2% |
| ov_SC06_008 | 2249 | 1702 | 0 | 2255 | 2542 | 88.7% |
| ov_SC06_010 | 2232 | 1700 | 0 | 2237 | 2517 | 88.9% |
| ov_SC06_011 | 2228 | 1700 | 0 | 2232 | 2468 | 90.4% |
| ov_SC06_013 | 2221 | 1700 | 0 | 2222 | 2425 | 91.6% |
| ov_SC06_014 | 2227 | 1700 | 0 | 2229 | 2453 | 90.9% |
| ov_SC06_015 | 2224 | 1700 | 0 | 2224 | 2421 | 91.9% |
| ov_SC06_016 | 2243 | 1700 | 0 | 2245 | 2549 | 88.1% |
| ov_SC06_018 | 2238 | 1703 | 0 | 2245 | 2665 | 84.2% |
| ov_SC06_020 | 2231 | 1702 | 0 | 2232 | 2518 | 88.6% |
| ov_SC06_022 | 2245 | 1700 | 0 | 2253 | 2642 | 85.3% |
| ov_SC06_024 | 2252 | 1700 | 0 | 2258 | 2667 | 84.7% |
| ov_SC06_025 | 2242 | 1700 | 0 | 2247 | 2572 | 87.4% |
| ov_SC06_027 | 2210 | 1700 | 0 | 2211 | 2408 | 91.8% |
| ov_SC06_029 | 2268 | 1700 | 0 | 2280 | 2662 | 85.6% |
| ov_SC06_030 | 2223 | 1700 | 0 | 2223 | 2456 | 90.5% |
| ov_SC06_032 | 2235 | 1701 | 0 | 2242 | 2658 | 84.3% |
| ov_SC06_033 | 2238 | 1701 | 0 | 2245 | 2631 | 85.3% |
| ov_SC07_000 | 2238 | 1704 | 0 | 2240 | 2521 | 88.9% |
| ov_SC07_001 | 2223 | 1700 | 0 | 2225 | 2454 | 90.7% |
| ov_SC07_002 | 2250 | 1700 | 0 | 2254 | 2579 | 87.4% |
| ov_SC07_006 | 1992 | 1548 | 0 | 2072 | 2456 | 84.4% |
| ov_SC07_007 | 2009 | 1549 | 0 | 2093 | 2614 | 80.1% |
| ov_SC07_008 | 2207 | 1700 | 0 | 2207 | 2386 | 92.5% |
| ov_SC07_009 | 2217 | 1700 | 0 | 2219 | 2430 | 91.3% |
| ov_SC07_010 | 2020 | 1549 | 0 | 2103 | 2526 | 83.3% |
| ov_SC07_011 | 2008 | 1548 | 0 | 2088 | 2450 | 85.2% |
| ov_SC01_000 | 2220 | 1706 | 0 | 2220 | 2403 | 92.4% |
| ov_SC01_001 | 2220 | 1706 | 0 | 2222 | 2466 | 90.1% |
| ov_SC01_004 | 2212 | 1697 | 0 | 2213 | 2414 | 91.7% |
| ov_SC01_005 | 2241 | 1719 | 0 | 2241 | 2503 | 89.5% |
| ov_SC01_006 | 2241 | 1719 | 0 | 2241 | 2503 | 89.5% |
| ov_SC01_008 | 2212 | 1697 | 0 | 2214 | 2426 | 91.3% |
| ov_SC01_009 | 2237 | 1698 | 0 | 2238 | 2507 | 89.3% |
| ov_SC01_074 | 2215 | 1698 | 0 | 2217 | 2425 | 91.4% |
| ov_SC01_077 | 2405 | 1666 | 0 | 2407 | 2585 | 93.1% |
| ov_SC01_080 | 2254 | 1701 | 0 | 2254 | 2512 | 89.7% |
| ov_SC01_084 | 2261 | 1701 | 0 | 2266 | 2579 | 87.9% |
| ov_SC02_000 | 2314 | 1736 | 0 | 2314 | 2683 | 86.2% |
| ov_SC02_003 | 2314 | 1736 | 0 | 2314 | 2683 | 86.2% |
| ov_SC02_004 | 2220 | 1701 | 0 | 2220 | 2401 | 92.5% |
| ov_SC02_005 | 2327 | 1697 | 0 | 2337 | 2927 | 79.8% |
| ov_SC02_011 | 2338 | 1704 | 0 | 2349 | 2893 | 81.2% |
| ov_SC02_015 | 2221 | 1703 | 0 | 2221 | 2414 | 92.0% |
| ov_SC02_016 | 2252 | 1703 | 0 | 2255 | 2545 | 88.6% |
| ov_SC02_017 | 2298 | 1703 | 0 | 2306 | 2732 | 84.4% |
| ov_SC02_021 | 2227 | 1703 | 0 | 2227 | 2437 | 91.4% |
| ov_SC02_026 | 2242 | 1701 | 0 | 2248 | 2572 | 87.4% |
| ov_SC02_027 | 2265 | 1701 | 0 | 2274 | 2692 | 84.5% |
| ov_SC02_028 | 2266 | 1701 | 0 | 2276 | 2701 | 84.3% |
| ov_SC02_031 | 2252 | 1702 | 0 | 2257 | 2562 | 88.1% |
| ov_SC02_035 | 2228 | 1701 | 0 | 2231 | 2520 | 88.5% |
| ov_SC02_039 | 2213 | 1701 | 0 | 2213 | 2416 | 91.6% |
| ov_SC02_041 | 2248 | 1701 | 0 | 2251 | 2561 | 87.9% |
| ov_SC03_001 | 2349 | 1702 | 0 | 2366 | 2869 | 82.5% |
| ov_SC03_002 | 2275 | 1706 | 0 | 2290 | 2631 | 87.0% |
| ov_SC03_003 | 2223 | 1701 | 0 | 2224 | 2423 | 91.8% |
| ov_SC03_006 | 2288 | 1707 | 0 | 2297 | 2767 | 83.0% |
| ov_SC03_007 | 2268 | 1701 | 0 | 2272 | 2623 | 86.6% |
| ov_SC03_010 | 2229 | 1701 | 0 | 2229 | 2470 | 90.2% |
| ov_SC03_011 | 2237 | 1701 | 0 | 2243 | 2527 | 88.8% |
| ov_SC03_012 | 2215 | 1701 | 0 | 2216 | 2406 | 92.1% |
| ov_SC03_013 | 2234 | 1701 | 0 | 2234 | 2492 | 89.6% |
| ov_SC03_014 | 2293 | 1726 | 0 | 2293 | 2685 | 85.4% |
| ov_SC03_015 | 2293 | 1726 | 0 | 2293 | 2685 | 85.4% |
| ov_SC03_023 | 2221 | 1701 | 0 | 2222 | 2435 | 91.3% |
| ov_SC03_024 | 2280 | 1706 | 0 | 2287 | 2641 | 86.6% |
| ov_SC03_028 | 2261 | 1701 | 0 | 2265 | 2664 | 85.0% |
| ov_SC03_029 | 2263 | 1703 | 0 | 2273 | 2644 | 86.0% |
| ov_SC03_030 | 2236 | 1706 | 0 | 2238 | 2495 | 89.7% |
| ov_SC03_031 | 2233 | 1701 | 0 | 2236 | 2514 | 88.9% |
| ov_SC03_089 | 2247 | 1701 | 0 | 2254 | 2581 | 87.3% |
| ov_SC03_090 | 2249 | 1701 | 0 | 2257 | 2625 | 86.0% |
| ov_SC03_091 | 2252 | 1701 | 0 | 2260 | 2640 | 85.6% |
| ov_SC03_092 | 2263 | 1701 | 0 | 2275 | 2587 | 87.9% |
| ov_SC03_093 | 2239 | 1701 | 0 | 2243 | 2563 | 87.5% |
| ov_SC03_094 | 2234 | 1701 | 0 | 2240 | 2577 | 86.9% |
| ov_SC03_095 | 2224 | 1701 | 0 | 2227 | 2474 | 90.0% |
| ov_SC03_096 | 2224 | 1701 | 0 | 2227 | 2465 | 90.3% |
| ov_SC03_097 | 2253 | 1701 | 0 | 2260 | 2603 | 86.8% |
| ov_SC03_098 | 2232 | 1701 | 0 | 2235 | 2541 | 88.0% |
| ov_SC03_099 | 2225 | 1701 | 0 | 2228 | 2504 | 89.0% |
| ov_SC03_100 | 2235 | 1701 | 0 | 2239 | 2539 | 88.2% |
| ov_SC03_101 | 2235 | 1701 | 0 | 2239 | 2528 | 88.6% |
| ov_SC03_102 | 2227 | 1701 | 0 | 2230 | 2492 | 89.5% |
| ov_SC03_103 | 2230 | 1703 | 0 | 2233 | 2510 | 89.0% |
| ov_SC03_104 | 2255 | 1701 | 0 | 2262 | 2616 | 86.5% |
| ov_SC03_105 | 2249 | 1701 | 0 | 2256 | 2597 | 86.9% |
| ov_SC03_108 | 2217 | 1701 | 0 | 2217 | 2443 | 90.7% |
| ov_SC03_109 | 2218 | 1701 | 0 | 2220 | 2424 | 91.6% |
| ov_SC03_110 | 2223 | 1701 | 0 | 2223 | 2468 | 90.1% |
| ov_SC03_111 | 2235 | 1701 | 0 | 2238 | 2510 | 89.2% |
| ov_SC03_112 | 2228 | 1703 | 0 | 2230 | 2526 | 88.3% |
| ov_SC03_113 | 2221 | 1703 | 0 | 2224 | 2468 | 90.1% |
| ov_SC03_114 | 2211 | 1701 | 0 | 2213 | 2415 | 91.6% |
| ov_SC03_115 | 2228 | 1701 | 0 | 2230 | 2472 | 90.2% |
| ov_SC03_116 | 2218 | 1701 | 0 | 2221 | 2438 | 91.1% |
| ov_SC03_117 | 2246 | 1701 | 0 | 2252 | 2557 | 88.1% |
| ov_SC03_118 | 2293 | 1725 | 0 | 2294 | 2685 | 85.4% |
| ov_SC03_119 | 2292 | 1725 | 0 | 2293 | 2685 | 85.4% |
| ov_SC03_121 | 2225 | 1701 | 0 | 2228 | 2459 | 90.6% |
| ov_SC03_124 | 2305 | 1697 | 0 | 2325 | 2741 | 84.8% |
| ov_SC03_125 | 2262 | 1701 | 0 | 2274 | 2588 | 87.9% |
| ov_SC03_126 | 2224 | 1703 | 0 | 2224 | 2423 | 91.8% |
| ov_SC04_000 | 2240 | 1705 | 0 | 2249 | 2546 | 88.3% |
| ov_SC04_002 | 2256 | 1701 | 0 | 2260 | 2636 | 85.7% |
| ov_SC04_003 | 2234 | 1701 | 0 | 2238 | 2502 | 89.4% |
| ov_SC04_004 | 2243 | 1701 | 0 | 2245 | 2558 | 87.8% |
| ov_SC04_005 | 2251 | 1701 | 0 | 2257 | 2612 | 86.4% |
| ov_SC04_006 | 2224 | 1701 | 0 | 2226 | 2454 | 90.7% |
| ov_SC04_007 | 2244 | 1701 | 0 | 2248 | 2569 | 87.5% |
| ov_SC04_008 | 2214 | 1701 | 0 | 2214 | 2415 | 91.7% |
| ov_SC04_009 | 2230 | 1701 | 0 | 2233 | 2441 | 91.5% |
| ov_SC04_010 | 2221 | 1701 | 0 | 2222 | 2419 | 91.9% |
| ov_SC04_011 | 2277 | 1701 | 0 | 2283 | 2803 | 81.4% |
| ov_SC04_012 | 2217 | 1701 | 0 | 2218 | 2420 | 91.7% |
| ov_SC04_015 | 2282 | 1702 | 0 | 2293 | 2611 | 87.8% |
| ov_SC04_016 | 2220 | 1701 | 0 | 2222 | 2440 | 91.1% |
| ov_SC04_018 | 2332 | 1736 | 0 | 2332 | 2857 | 81.6% |
| ov_SC04_019 | 2338 | 1736 | 0 | 2338 | 2857 | 81.8% |
| ov_SC04_020 | 2249 | 1701 | 0 | 2261 | 2567 | 88.1% |
| ov_SC04_021 | 2225 | 1703 | 0 | 2225 | 2423 | 91.8% |
| ov_SC05_000 | 2221 | 1705 | 0 | 2224 | 2422 | 91.8% |
| ov_SC05_001 | 2244 | 1701 | 0 | 2249 | 2574 | 87.4% |
| ov_SC05_002 | 2226 | 1701 | 0 | 2229 | 2442 | 91.3% |
| ov_SC05_003 | 2223 | 1701 | 0 | 2224 | 2481 | 89.6% |
| ov_SC05_004 | 2218 | 1701 | 0 | 2220 | 2464 | 90.1% |
| ov_SC05_005 | 2225 | 1701 | 0 | 2226 | 2491 | 89.4% |
| ov_SC05_006 | 2217 | 1701 | 0 | 2217 | 2430 | 91.2% |
| ov_SC05_007 | 2227 | 1701 | 0 | 2232 | 2482 | 89.9% |
| ov_SC05_008 | 2244 | 1701 | 0 | 2246 | 2543 | 88.3% |
| ov_SC05_009 | 2222 | 1701 | 0 | 2226 | 2438 | 91.3% |
| ov_SC05_010 | 2248 | 1701 | 0 | 2252 | 2588 | 87.0% |
| ov_SC05_011 | 2220 | 1701 | 0 | 2221 | 2409 | 92.2% |
| ov_SC05_017 | 2328 | 1698 | 0 | 2339 | 2842 | 82.3% |
| ov_SC05_018 | 2275 | 1701 | 0 | 2288 | 2673 | 85.6% |
| ov_SC05_019 | 2225 | 1703 | 0 | 2225 | 2423 | 91.8% |
| ov_SC06_000 | 2287 | 1710 | 0 | 2290 | 2691 | 85.1% |
| ov_SC06_006 | 2240 | 1701 | 0 | 2241 | 2511 | 89.2% |
| ov_SC06_008 | 2250 | 1703 | 0 | 2256 | 2542 | 88.7% |
| ov_SC06_010 | 2233 | 1701 | 0 | 2238 | 2517 | 88.9% |
| ov_SC06_011 | 2229 | 1701 | 0 | 2233 | 2468 | 90.5% |
| ov_SC06_013 | 2222 | 1701 | 0 | 2223 | 2425 | 91.7% |
| ov_SC06_014 | 2228 | 1701 | 0 | 2230 | 2453 | 90.9% |
| ov_SC06_015 | 2225 | 1701 | 0 | 2225 | 2421 | 91.9% |
| ov_SC06_016 | 2243 | 1701 | 0 | 2245 | 2549 | 88.1% |
| ov_SC06_018 | 2239 | 1704 | 0 | 2246 | 2665 | 84.3% |
| ov_SC06_020 | 2232 | 1703 | 0 | 2233 | 2518 | 88.7% |
| ov_SC06_022 | 2246 | 1701 | 0 | 2254 | 2642 | 85.3% |
| ov_SC06_024 | 2252 | 1701 | 0 | 2258 | 2667 | 84.7% |
| ov_SC06_025 | 2243 | 1701 | 0 | 2248 | 2572 | 87.4% |
| ov_SC06_027 | 2211 | 1701 | 0 | 2212 | 2408 | 91.9% |
| ov_SC06_029 | 2269 | 1701 | 0 | 2281 | 2662 | 85.7% |
| ov_SC06_030 | 2224 | 1701 | 0 | 2224 | 2456 | 90.6% |
| ov_SC06_032 | 2236 | 1702 | 0 | 2243 | 2658 | 84.4% |
| ov_SC06_033 | 2239 | 1702 | 0 | 2246 | 2631 | 85.4% |
| ov_SC07_000 | 2239 | 1705 | 0 | 2241 | 2521 | 88.9% |
| ov_SC07_001 | 2224 | 1701 | 0 | 2226 | 2454 | 90.7% |
| ov_SC07_002 | 2251 | 1701 | 0 | 2255 | 2579 | 87.4% |
| ov_SC07_006 | 1993 | 1549 | 0 | 2073 | 2456 | 84.4% |
| ov_SC07_007 | 2010 | 1550 | 0 | 2094 | 2614 | 80.1% |
| ov_SC07_008 | 2208 | 1701 | 0 | 2208 | 2386 | 92.5% |
| ov_SC07_009 | 2218 | 1701 | 0 | 2220 | 2430 | 91.4% |
| ov_SC07_010 | 2021 | 1550 | 0 | 2104 | 2526 | 83.3% |
| ov_SC07_011 | 2009 | 1549 | 0 | 2089 | 2450 | 85.3% |
phase-ends/CURRENT_PHASE.md
+18 -3
View File
@@ -67,9 +67,24 @@ arithmetic scales (R14/R35).** Every prior "structural wall" (B2, SC07, pin-cras
`residual_class.py` + `autopsy.py` + `corpus.is_o0`; grinder targeting wired (1,303→78). FINDING: only
**75/972 (7.7%)** of the grinder-admissible backlog is permuter-shaped — the problem was TARGETING, not
a missing transform. R22 140/140. Cookbook §60 + decision-log + SETUP.
- [ ] ▶ **Task 13B (RE-SCOPED) — run the DIRECTED permuter over the 75, then autopsy what actually plateaus**
— the LLM batch autopsy is deferred until genuine plateaus exist (R35: do not diagnose a search that
never ran under correct targeting). Then Task 14 (gate_stage ladder §57/§59 — priced by the 1/12 probe).
- [x] **Task 13B — directed permuter run + the plateau autopsy [Max]** ✅ 2026-07-21
Ran the targeted grinder over all 77 permuter-bucket candidates: **24 wins → 21 banked** (27%), 3
won-but-gate-rejected as PLUMBING. `func_80141B90` propagated **×138**. Then the `length` expansion:
probe = 20 targets, **1 win** — `tail` 1/6, `partial` **0/12**. **AUTOPSY VERDICT: ZERO
missing-transforms.** The partial plateaus are WRONG DRAFTS (byte-read: `(u32)(x^1)`→`xori` vs the
target's `sltiu`=`!x`; a draft storing to `arg0+8` vs a target storing to a global). So the §7
"extend the mutation set" bucket came back EMPTY for this class, and the permanent win was the
opposite move — a tighter ADMISSION rule: `_drift_route` now needs `|Δ|≤2` AND `explains=="tail"`
(length pool 339→**34**, permuter bucket 389→**84**), plus a proportional `SIZE-MISMATCH` test.
Cookbook **§60a/§60b**. `permuter_weights` needs no extension.
- [ ] ▶ **Task 14 — enrich `gate_stage`'s ladder (§57/§59) so more wins auto-integrate + grinder auto-R22
[Max]** — the permuter-automation program's final stage (Tasks 12→13→14). Justified as the SHARED
BANKING SPINE, not as a bucket: today 3 of 24 permuter wins were lost to plumbing (12.5% of work
already paid for in CPU), the 12-draft probe banked 1/12, and crack-wave-4's 6 -O2 cores ×138 were
explicitly deferred to this ladder. **Corrected ceiling for the integration bucket alone: 0.84pp**
(NOT the 3.2pp I first quoted — see the valuation correction below). Start by splicing 6–10 failures
individually and reading real cc1/ld stderr: the shared `conflicting types for built-in function`
label is a §58 red-herring, so the causes are UNDIAGNOSED.
- [ ] **Task 2 — Act on the verdict [Max/xHigh]** — **branch = 2a (tooling).** Build the -O0-cluster
split rollout (adapt `rollout_whale_o0.py` → per-overlay `[0x13410,c,<ov>_o0]` carve, -O0 Makefile
rule) + teach `family_sweep --hseq` per-member opt-level awareness; WHOLE-BINARY gate one overlay
tools/residual_class.py
+21 -10
View File
@@ -207,12 +207,12 @@ def classify_streams(mine, tgt):
return _v("LENGTH-DRIFT", closeness, nm, nt,
{"delta": d, "at": k, "explains": "tail",
"mine_extra": [_show(mine, k + t) for t in range(min(d, 4))]},
sig="LENGTH-DRIFT/+%d" % d, **_drift_route(d))
sig="LENGTH-DRIFT/+%d" % d, **_drift_route(d, "tail"))
if d < 0 and all(eqm(k + t, k - d + t) for t in range(nm - k)):
return _v("LENGTH-DRIFT", closeness, nm, nt,
{"delta": d, "at": k, "explains": "tail",
"tgt_extra": [_show(tgt, k + t) for t in range(min(-d, 4))]},
sig="LENGTH-DRIFT/%d" % d, **_drift_route(d))
sig="LENGTH-DRIFT/%d" % d, **_drift_route(d, "tail"))
# No single shift point explains it. Now the SIZE ratio decides between two very different
# situations that a scalar `closeness` renders identical:
# a draft 1-2 instructions off -> a real near-miss; a seed tweak or a known idiom closes it
@@ -222,14 +222,18 @@ def classify_streams(mine, tgt):
# second wants a re-draft — and a corpus full of the second, ranked by closeness, is how a
# queue of NEVER-PROPERLY-ATTEMPTED functions comes to look like a wall of hard ones.
# Evidence first (the shift test above), ratio only as the fallback discriminator.
if abs(d) > max(2, 0.15 * nt):
# `max(2, 0.15*nt)` alone is far too permissive for a TINY target: a 2-instruction draft
# against a 4-instruction target is |d|=2 and reads as a near-miss, when it is a different
# function (byte-witnessed: func_8017FF90 stored to arg0+8, the target to a global). Add a
# PROPORTIONAL test so smallness cannot disguise a wholesale mismatch.
if abs(d) > max(2, 0.15 * nt) or abs(d) >= 0.5 * nt:
return _v("SIZE-MISMATCH", closeness, nm, nt,
{"delta": d, "at": k, "ratio": round(nm / float(nt), 2),
"note": "draft is a different-sized function — re-draft, do not seed-tweak"},
sig="SIZE-MISMATCH/%s" % ("short" if d < 0 else "long"))
return _v("LENGTH-DRIFT", closeness, nm, nt,
{"delta": d, "at": k, "explains": "partial"},
sig="LENGTH-DRIFT/%d?" % d, **_drift_route(d))
sig="LENGTH-DRIFT/%d?" % d, **_drift_route(d, "partial"))
if closeness > 8:
# equal lengths: an insert+delete pair re-aligns the middle. Accept a shift only if it
# explains most of the run (a coincidental partial alignment must not become a verdict).
@@ -438,12 +442,19 @@ def _family(a, b):
return None
def _drift_route(d):
"""A LENGTH-DRIFT of +-1..2 is permuter-shaped (one local add/drop of an instruction); anything
larger is a wrong-shaped draft. Phase-29 measurement: |d|<=2 is 339 fns / 472k instruction-
weighted member-instances (~3.6pp of the fleet) — 10x the entire pre-existing permuter bucket,
which is why this magnitude split is worth having at all."""
return {"profile": "length", "bucket": "permuter"} if abs(d) <= 2 else {}
def _drift_route(d, explains="partial"):
"""Permuter-shaped ONLY when the drift is small AND a single shift point explains the whole
tail (`explains == "tail"`): that is the shape "everything matches except one added/dropped
instruction", which perm_temp_for_expr / perm_expand_expr can actually reach.
MEASURED (Phase-29 Task-13B, 20-target probe): `tail` converted 1/6, `partial` converted 0/12.
Autopsy of the partial plateaus found them to be WRONG DRAFTS, not missing transforms — e.g. a
draft writing `(u32)(x ^ 1)` (emits `xori`) against a target doing `!x` (emits `sltiu rd,rs,1`),
and a draft storing to `arg0+8` against a target storing to a GLOBAL. No local mutation crosses
either gap. So a `partial` drift is seed-structural: it wants a re-draft, and admitting it to
the permuter spends CPU that can never pay (the same waste Task-13A's targeting reclaimed —
this is the second, finer cut of the same knife)."""
return {"profile": "length", "bucket": "permuter"} if (abs(d) <= 2 and explains == "tail") else {}
def _v(klass, closeness, nm, nt, detail, sig=None, profile=_KEEP, bucket=_KEEP):