Drew T 26ba449684 perf(tools-health): parallelise the sig targets; fix a latent probe-file race; MEASURE the real cost ...

Drew asked why `make tools-health` runs 15+ min. Measured per step rather than
guessed (I guessed wrong twice first, and both are recorded in the comments):

  sig-overlays  ~52s serial  -> 3.9s wall / 51.8s user  (xargs -P$(JOBS), 32 cores)
  sig-modules   0s   sig-resident 0s   audit-corpus 17s
  audit-cdecl   >9 MINUTES  <-- the actual bottleneck, and NOT the gcc probes:
                the `[gcc] N distinct declarations` line never printed inside a
                10-minute run, so not one cc1 call had happened. `tu_statements`
                over 4,168 TUs is ~787s single-core, all of it before the probes.

SHIPPED
  * sig-overlays / sig-modules: xargs -P$(JOBS), same pattern extract-all and
    check-all already use in this file. sig_image has exactly one write path
    (its own per-alias .jsonl), verified before fanning out. NEGATIVE CONTROL:
    141/141 sig files BYTE-IDENTICAL to the serial output. Also adds the failure
    detection the serial loops never had -- a sig_image crash used to vanish (R32).
  * cdecl._gcc_probe: `probe_{tag}.c` was ONE FIXED FILENAME PER TAG, correct only
    while _sift is serial. Now unique per call, so concurrent probes cannot
    overwrite each other's source between write and compile and return a verdict
    about another chunk's declarations.
  * cdecl._sift: threads over chunks + over the bisection probes (gcc is a
    subprocess, so the GIL is released), results written back BY INDEX so the
    output stays deterministic. A/B on --limit 6: IDENTICAL verdicts (829/829).

NOT SHIPPED, and the measurement is left in the code
  A ProcessPoolExecutor over the collection phase was tried and REVERTED: 12 TUs
  yield 32,352 statements, so the full pass ships ~11M strings through IPC and the
  pickling costs more than the parse it saves. The fix is to dedupe/filter INSIDE
  the worker or memoise per-TU by content hash -- left measured, not guessed.

2026-09-01 20:10:51 -06:00

..

asm-differ @ 2ad4a4a4f7

feat(phase-4): matching-build toolchain (WSL) — make check-env green

2026-06-14 03:56:07 -06:00

bfm_extract

feat(phase-3.5): prototype research spike — go/no-go (protos don't accelerate Gen1 matching)

2026-06-14 03:19:23 -06:00

bin

chore(gen2-prep): Gen2 roadmap + private-repo backup hardening + SETUP/submodule fixes

2026-06-15 17:15:31 -06:00

brave-CUE

feat(phase-2): deterministic disc/.CD/PAC/LZSS extraction pipeline

2026-06-13 19:47:42 -06:00

decomp-permuter @ b44b062226

feat(phase-4): matching-build toolchain (WSL) — make check-env green

2026-06-14 03:56:07 -06:00

ghidra_scripts

chore(phase-30 S45 II.2): retirements (R33) + SETUP module recipe

2026-08-06 13:34:50 -06:00

lanes

feat(resolver): the integration-resolver lane — zero-token re-judging of the ledgers' shape-correct stock at the real TU (P31 S61 T10+/S61-1)

2026-08-25 23:50:26 -06:00

m2c @ 4266cc2817

feat(phase-4): matching-build toolchain (WSL) — make check-env green

2026-06-14 03:56:07 -06:00

maspsx @ 874855c53f

feat(phase-4): matching-build toolchain (WSL) — make check-env green

2026-06-14 03:56:07 -06:00

oracle

feat(phase-29): T46 — reg_renumber-swap oracle built + validated; it REFUTES the func_80176734 framing

2026-07-28 14:53:47 -06:00

permuter

feat(phase-24): T3 — permuter setup fixes (pins/GTE-asm, typedefs, -O0); flagship closed

2026-07-02 23:32:38 -06:00

wave

feat(phase-30 S49): the A-prop word-diff card + aprop_wave — 56 banked from the >=16 head (§170)

2026-08-13 17:44:20 -06:00

workflows

feat(triage): the triage ladder — built, wired, and acceptance-green

2026-08-31 23:53:50 -06:00

ab_score.py

feat(phase-22): cheap-tier A/B experiment harness + findings

2026-06-29 12:52:39 -06:00

api_agent.py

fix(api_agent): prior_draft accepts a warm-start body only when its symbols overlap the target .s (law 1c) — 22/60 T4 agents had received another overlay's same-named function; fleet ledger regenerated (2,380 stubs) (P31 S62)

2026-08-26 19:02:07 -06:00

api_draft.py

feat(phase-31): S57 external-model bake-off — 4 tool fixes, §205/§206, api_agent harness

2026-08-23 00:12:07 -06:00

api_rate.py

docs+rules(S58): R42 commit-banked-work-immediately, R43 refuse-unsupported-input

2026-08-23 12:59:59 -06:00

aprop_autodraft.py

feat(resolver/aprop): decl-from-use wired into the A-prop drafter; TU-BROKEN classification; dup-def→extern candidates (P31 S61 follow-up, all zero-token)

2026-08-26 01:14:36 -06:00

aprop_symfix.py

fix(phase-31): aprop_symfix survives curated (non-hex) symbol names

2026-08-15 01:32:34 -06:00

asm_verbatim.py

fix(pgate): link the signature registry into worktrees — every worktree CARVE-REFUSED was an artifact

2026-09-01 14:09:38 -06:00

atlas_features.py

feat(phase-31 T5): THE FRONTIER ATLAS — 5,139 lever-labeled crack groups over all 12,058 open fns

2026-08-14 18:06:46 -06:00

atlas.py

perf(throughput): unblock the main lane, both gate logs, deeper wave overlap, atomic atlas

2026-08-25 11:28:51 -06:00

audit_binaries.py

fix(phase-30 S44 I.1c): the R36 citizenship gate is no longer blind to non-ov_ binary classes

2026-08-06 10:55:38 -06:00

audit_digest.py

fix(phase-30 S1e): the distinct-code "regression" was a STALE DIGEST — alias lever ungated

2026-08-04 21:48:57 -06:00

audit_frontier.py

fix(phase-30 S44 I.1d): the module class reaches every enumerating consumer

2026-08-06 10:57:17 -06:00

audit_header_sigs.py

feat(phase-29): T73 — items 1+2: ARITY class resolved, audit learns §113; DECLS is the last value

2026-07-29 00:42:40 -06:00

audit_text_sources.py

fix(phase-30): 137 tracked C sources held a raw NUL that made grep SILENTLY SKIP them

2026-07-31 14:18:11 -06:00

auto_driver.py

docs(phase-16): PIVOT — m2c+permuter won't crack the loose-typed core; new plan + harness fixes

2026-06-19 10:05:58 -06:00

auto_status.sh

docs(phase-21): automation runbook + status tool (grinder running; worker = remote-driven)

2026-06-21 13:12:04 -06:00

auto_stop.sh

feat(phase-16): S5 driver+supervisor+safe-exit + NULL fix + sig_unify pipeline

2026-06-18 23:15:31 -06:00

auto_supervisor.sh

feat(phase-21): T4/T5/T6 — worker validated (func_80164530 x134), grinder, orchestrator

2026-06-21 13:06:34 -06:00

autopsy.py

fix(phase-29): T54 — correct the ADDRESSING route, and fix the reason changing it was inert

2026-07-28 19:34:00 -06:00

backlog.py

fix(phase-30 S44 I.1d): the module class reaches every enumerating consumer

2026-08-06 10:57:17 -06:00

bank_exemplar.py

fix(phase-26a): A9e — reconcile_tu already wired into bank_exemplar (A3d); document the ladder

2026-07-14 21:25:19 -06:00

bisect_slate.py

feat(phase-31): wave Q — 40 main functions banked; + fragment_check.py and bisect_slate.py

2026-08-16 17:21:10 -06:00

blast_radius.py

feat(phase-29): tools/blast_radius.py — MEASURE the write set, enforce the §63 tier (step 1)

2026-07-26 23:08:07 -06:00

blocker_probe.py

fix(rtu_match/blocker_probe): main was structurally unprobeable — pass the TU path and asm subdir, never reconstruct them

2026-08-31 16:34:01 -06:00

build_engine_types.py

feat(phase-29): broad §20 type-lift lands — 154 types fleet-wide, R22 140/140 (§64)

2026-07-23 19:54:19 -06:00

build_fuel_manifest.py

fix(phase-26a): A3 — target selection was blind to 91.6% of the remaining work; now derived

2026-07-14 09:22:31 -06:00

build_wave_args.py

feat(phase-29): ov_SC07_006 reach-138 batch-1 propagation + live-count re-scope (SESSION-13)

2026-07-23 11:43:28 -06:00

build_wave_atlas.py

fix(wave): cards keyed by (binary,fn) — t5a received 48/48 ANOTHER binary's card (R48); t5_cards.py builds the target's OWN fuel (tu_ref 88% / decl_prior 98%) via wave_card_fuel.py extracted from build_wave_atlas (R33); t5_bank.sh recovers un-banked drafts via fix_tu_ret_decls (2/2 on t5a); distill workflow generalized (P31 S63 T5)

2026-08-26 20:19:41 -06:00

build_wave.py

feat(phase-31): wave C banked 32 (R22 213/213) + tools/build_wave.py; 3 instrument defects fixed

2026-08-14 23:44:03 -06:00

bulk_harvest.py

feat(phase-30 S45p7): Stage 1 complete — shared-state RW lock + 15 more banks (wave-3 revived)

2026-08-07 18:57:10 -06:00

burndown.py

fix(phase-29): the FLOOR VERDICT was an artifact of snapshot frequency, not yield decay

2026-07-22 15:24:59 -06:00

campaign_status.py

feat(tools): T5pre — main_queue_rebuild.py derives main's open set from progress.classify() (147 = 144 stubs + 3 NM; 959 LINKED + 2 blobs excluded) and attaches the newest draft per fn (136 queued, 11 never-drafted listed for the main wave; the old queue was 156/166 stale); campaign_status quotes the honest main denominator (P31 S62)

2026-08-26 15:02:15 -06:00

canon_draft_decls.py

feat(phase-15): T6d — deterministic decl-canonicalization recovers 19 more shared fns; fleet -> 31.22%

2026-06-17 00:30:45 -06:00

canon_resident_calls.py

feat(phase-19): T2 recovery tooling — canon_resident_calls + no-proto caller fix (3 matches)

2026-06-20 17:12:00 -06:00

canon_sig_reconcile.py

fix(phase-26a): A9a — canon_sig_reconcile sees fn-ptr dispatch tables (cdecl supplement)

2026-07-14 19:59:56 -06:00

cast_call_sites.py

fix(phase-30 S40): cast_call_sites read a RETURN as a prototype and deleted it — 0/39 sweep becomes 18/39

2026-08-05 11:05:35 -06:00

cast_self_callers.py

fix(undo-journal): REFUSE an ambiguous restore instead of silently swapping decls (§403)

2026-09-01 17:36:57 -06:00

cc1_dumps.sh

feat(phase-30 S50): func_8017C294 frame residual decoded to the instruction level (§172) — crack REFUTED across the reachable space, wall documented

2026-08-14 12:08:50 -06:00

cdecl.py

perf(tools-health): parallelise the sig targets; fix a latent probe-file race; MEASURE the real cost

2026-09-01 20:10:51 -06:00

cdtrace.py

feat(phase-30 S45p6): tools/cdtrace.py — a runtime CD-load oracle; 7 routing-table addresses confirmed live

2026-08-07 20:36:04 -06:00

claude_wave_packs.py

feat(packs): push the cookbook bucket into the pack — retrieval was the leak

2026-08-30 17:22:11 -06:00

conform_decls.py

feat(phase-30 S47-B): conform 8 declaration axes (~10,930 sites); 3 guard defects fixed; 213/213

2026-08-10 16:01:49 -06:00

cookbook_index.py

fix(flywheel): the index is the entry point, and a cited number that fails to grep is a LINE

2026-08-24 13:36:58 -06:00

corpus.py

feat(o0): md_MAIN_011 compiles -O0 — 21 functions / 4,321 ins unlocked, byte-proven

2026-08-24 12:43:40 -06:00

decl_from_use.py

feat(tools): decl_from_use — negative-controlled extern inference for the no-seed-decl class (Fable agent build, verified); SETUP.md rows for it + the baseline-red gate input (R21)

2026-08-26 00:47:38 -06:00

decl_prior.py

chore(phase-31): S56 checkpoint — 127 banked, wave AA built and staged (not launched)

2026-08-18 15:44:55 -06:00

decompile.py

feat(phase-6): compiler pinned + matching loop proven (14 matches) — LZSS to Phase 7

2026-06-14 14:47:01 -06:00

dedup_extend.py

feat(phase-30 S46-6): recovery ladder banks +6; wire gate_stage into dedup_extend

2026-08-10 10:43:44 -06:00

dedup_integrate.py

fix(phase-26a): A1 — dedup_integrate was a gate that could print a FALSE GREEN

2026-07-14 02:50:28 -06:00

dedup_propagate.py

feat(p31 s68): deferred propagation done honestly (2 banked) + seed_ref was offering DEAD TEXT

2026-08-31 15:59:01 -06:00

demacroize.py

feat(phase-29): T16.6/T16.7 — 14th bank via the generalized de-macroize; §65 distilled; R21 inventory debt cleared

2026-07-24 14:25:28 -06:00

diff_autopsy.sh

fix(tools): diff_autopsy.sh creates its scratch dir (the promoted copy lost the mkdir, so its snapshot cp failed and the restore left four TUs spliced), prints the first build errors and flags a stale cmp when the build fails (P31 S62)

2026-08-26 15:41:20 -06:00

diff_regions.py

feat(phase-29 T1): the swing number RESOLVED — (a) tooling, an -O0 compile-flag artifact (not a wall)

2026-07-16 12:07:37 -06:00

difficulty.py

feat(phase-27 T6): migrate difficulty + exemplar_miner off hand-lists/proxies (R33) — before T7

2026-07-15 18:20:55 -06:00

disc_audit.py

fix(phase-30 S43): audit-disc listed only L1's code — the real backlog is 78 payloads, not 39

2026-08-06 09:51:39 -06:00

distill_scan.py

fix(distill): a stale READY marker no longer blocks the lane forever

2026-08-25 00:49:24 -06:00

draft_prechecks.py

feat(phase-30 S50): the mechanical A-prop draft — 256 members banked with no agent in the loop

2026-08-14 07:42:19 -06:00

draw_waves.py

feat(draw): NO SONNET — opus <=150 ins, fable >150 (Drew, 2026-09-01)

2026-09-01 11:18:23 -06:00

dup_report.py

feat(phase-30 S45 L3-p3): SC02/9 = the Steam Knight boss module — decoded, captured, retro-verified, onboarded; parked = 5

2026-08-07 15:07:55 -06:00

eval_lora.py

feat(phase-22): eval_lora held-out scorer (validated) + train_lora --max-steps

2026-06-29 16:49:11 -06:00

exemplar_miner.py

feat(phase-27 T6): migrate difficulty + exemplar_miner off hand-lists/proxies (R33) — before T7

2026-07-15 18:20:55 -06:00

export_pairs.py

feat(phase-23): local GPU serving + prompt-fix + corpus-v3 (macro+struct mining)

2026-06-30 01:46:54 -06:00

family_align.py

feat(phase-31 T8): LEN+N lane — 587 near-misses routed; 345 wrong-drafts reclassified; detectors live

2026-08-14 19:54:57 -06:00

family_cousins.py

feat(phase-31 T9): warmstart feeder + weak-cards — the campaign lanes are armed

2026-08-14 19:59:21 -06:00

family_hseq.py

fix(phase-30 S47): family_hseq derives its own SCOPE, not just its own count; + a zero-crack glossary

2026-08-11 13:42:40 -06:00

family_manifest.py

fix(phase-26a): A3 — the endgame plan was 2.8x too big; the matched set is now DERIVED

2026-07-14 09:40:38 -06:00

family_remap.py

fix(family_remap): destination TU decls win over carried exemplar externs (§398)

2026-09-01 17:34:21 -06:00

family_sweep.py

perf: -j on the last three per-binary make build sites (family_sweep/twin_sweep, rollout_o0, restore_dropped_decls)

2026-08-31 14:24:03 -06:00

find_addr_refs.py

feat(phase-30 S46-2): the master IDXTAB/DESTPTR load map — and the tracker blind spot that hid it

2026-08-07 22:09:05 -06:00

fix_arity_callers.py

fix(undo-journal): REFUSE an ambiguous restore instead of silently swapping decls (§403)

2026-09-01 17:36:57 -06:00

fix_decl_mirror.py

feat(tools): fix_decl_mirror.py — the MIRROR of fix_tu_ret_decls (TU declares a VALUE return for a genuinely void definition; it SKIPped in every S63 wave). Narrowing the TU decl is NOT byte-neutral there (a caller reads $v0), so the repair is draft-side: define the body under a distinct identifier bound by __asm__("func_X") — the tree's own impl_* idiom. T0 blast radius. R39 negative control found and killed TWO real defects (block-scope decls: 856 false positives; sibling-TU decls: 580) and now passes 78,727/78,727 clean. Wired into t5_bank.sh (P31 S64)

2026-08-27 12:19:36 -06:00

fix_header_decl.py

feat(phase-29): fix_header_decl v2 (--reconcile-externs) + SESSION-13 cont checkpoint

2026-07-23 12:42:43 -06:00

fix_tu_ret_decls.py

fix(aprop): three more decl-layer defects, each traced to a wrong-name or wrong-scope lookup

2026-08-24 23:19:47 -06:00

format_finetune.py

feat(phase-23): local GPU serving + prompt-fix + corpus-v3 (macro+struct mining)

2026-06-30 01:46:54 -06:00

fragment_check.py

docs+tools(phase-31): cookbook §180d/§181/§182; fragment_check gains the mirror test (5.7% of main stubs); pregate_check scans masked text

2026-08-16 21:07:03 -06:00

gap_triage.py

fix(phase-31): gap_triage — threshold at 0.15, because a weak suggestion costs a reader's attention

2026-08-18 11:08:29 -06:00

gate_lane.py

fix(main-lane): the baseline was RED — guard every gate with a no-draft control (S59)

2026-08-24 19:08:47 -06:00

gate_main.py

fix(main-lane): the baseline was RED — guard every gate with a no-draft control (S59)

2026-08-24 19:08:47 -06:00

gate_stage.py

feat(tools): work_evidence — assert a tool ACTUALLY DID the work it reports

2026-09-01 18:59:04 -06:00

gate_wave.py

docs(p31 s67): wave-playbook (the CURRENT pipeline) + seed_ref fix + harvest §333-§338

2026-08-31 13:48:56 -06:00

gater_lane.py

fix(gater_lane): ledger a draft as gated only for a binary the gate actually EXAMINED

2026-09-01 17:37:49 -06:00

gen_engine_decls.py

docs(phase-16): PIVOT — m2c+permuter won't crack the loose-typed core; new plan + harness fixes

2026-06-19 10:05:58 -06:00

gen_harvest_targets.py

fix(phase-26): 10% of the canonical-callee oracle was silently missing (own-line-brace DEFINE macros)

2026-07-14 00:08:59 -06:00

gen_lib_subsegs.py

refactor(phase-9): T8a — de-default the EXE-curation helpers (thread --exe/--vram-base)

2026-06-15 20:55:28 -06:00

ghidra_import_raw.sh

feat(phase-10): T4 — resident Ghidra program (2nd binary) + symbol seed + sig

2026-06-15 23:25:00 -06:00

ghidra_import.sh

feat(phase-3.5): prototype research spike — go/no-go (protos don't accelerate Gen1 matching)

2026-06-14 03:19:23 -06:00

ghidra_mcp_start.sh

feat(phase-3.5): prototype research spike — go/no-go (protos don't accelerate Gen1 matching)

2026-06-14 03:19:23 -06:00

ghidra_mcp_stop.sh

feat(phase-3): file-loader & overlay-map RE + PCSX-Redux runtime oracle

2026-06-14 02:04:28 -06:00

ghidra_mcp_verify.sh

feat(phase-3): file-loader & overlay-map RE + PCSX-Redux runtime oracle

2026-06-14 02:04:28 -06:00

glm_parallel.sh

feat(phase-23): Fable5Max cracks §20 "unsteerable" giant func_8014EE14 ×134

2026-07-02 16:56:56 -06:00

glm_reconcile.py

feat(phase-23): T10.7 Option-3 — GLM reasons the def-side wall; +1 bank (func_80175184)

2026-07-01 21:03:08 -06:00

grinder.py

feat(phase-30): T4 — grinder ILS warm-restart wired in; SS119 posture audited + a stale doc hazard struck

2026-07-31 12:26:03 -06:00

harvest_verify.py

feat(tools): work_evidence — assert a tool ACTUALLY DID the work it reports

2026-09-01 18:59:04 -06:00

idiom_harvest.py

docs+rules(S58): R42 commit-banked-work-immediately, R43 refuse-unsupported-input

2026-08-23 12:59:59 -06:00

idiom_hunt.py

feat(phase-23): T10.8 idiom-hunt harness + calibration — GLM confirms our idioms, no NEW ones

2026-07-01 23:13:29 -06:00

idiom_loop.py

feat(phase-21): cont.7 — cheap close=0 recovery EXHAUSTED (byte-proven); fix --assess double inflation

2026-06-25 13:09:02 -06:00

idiom_serial.py

fix(serial lane): it was dead at turn 0 — a family-card file fed to a per-function consumer

2026-08-25 15:00:31 -06:00

idxtab_map.py

feat(phase-30 S46-2): the master IDXTAB/DESTPTR load map — and the tracker blind spot that hid it

2026-08-07 22:09:05 -06:00

inject_capped_externs.py

feat(phase-23): macro-extern-injection lever — 8 reach-134 fns freed ×134 (fleet 63.82→64.09%)

2026-06-30 11:15:11 -06:00

integration_resolver.py

feat(tools)+docs(phase-31): T1 — resolver declfix wiring; masked_diff/rtu_match compare internal-j targets (jrel; §195 premise refuted, cookbook §301); diff_autopsy.sh + stub_invariant_audit.py promoted; frontier-s61 Addendum 2 (step-1 falsifier fired: 54 carve-refused, 2 DIFF autopsied+banked, 1 plumbing); S62 log

2026-08-26 14:04:54 -06:00

interleave_check.py

fix(config): fleet clean sweep found 5 more drifted overlays that incremental gates had hidden — ov_SC03_002/ov_SC01_077/ov_SC04_019/ov_SC01_084 JTBL_INTERLEAVE orders regenerated from their yamls (interleave_check --fix), ov_SC05_010 jr_8017C8D0 pads 0,0 -> 0 (the 09:20 maintenance repair was blanket-reverted by the 10:21 ds1 wave commit); each byte-identical from a clean per-binary rebuild (P31 S62)

2026-08-26 14:34:32 -06:00

jr_isolate_all.py

feat(o0): md_MAIN_003 second carve — func_800D12D0 (657 ins) banked as real -O0 C

2026-08-31 19:46:10 -06:00

jr_isolate.py

feat(phase-26): §8b multi-jtbl same-subseg — contiguous MERGE (built) + isolation scaffold

2026-07-12 22:06:17 -06:00

jtbl_carve.py

feat(p31 s67): stranded-draft census + honest jtbl probe — the frontier's biggest class is carve plumbing

2026-08-31 01:46:16 -06:00

jtbl_family_bank.py

fix(tools): jtbl_family_bank reported "gate-fail" with no reason — the third missing-payload defect today

2026-08-04 19:31:33 -06:00

jtbl_lane.py

feat(jtbl): the carve→draft→bank pipeline is automated at the gate — byte-proven ×3, one per class (S59)

2026-08-24 12:31:14 -06:00

jtbl_pads_fix.py

fix(gate/tools): baseline-RED skip in the gate, jtbl_pads_fix un-broken three ways, mk_write parse guard, shadow join for 0-bank waves (P31 S61 T10+/S61-10)

2026-08-26 00:29:05 -06:00

jtbl_rodata_pads.py

fix(tools): jtbl_rodata_pads --derive — empty spec when the TU has no C tables (never refuses what it need not pad), sub-4-byte all-zero gaps before an anchor accepted as assembler alignment, .float/.double sized; 70/70 module binaries byte-identical through the derive stage; T3a log (P31 S62)

2026-08-26 14:50:12 -06:00

lane_inflight.py

fix(r22 guard): record liveness, do not infer it — my mtime heuristic failed BOTH ways

2026-08-31 18:59:50 -06:00

ld_interleave.py

feat(phase-26): §8 multi-jtbl --order carve + family-1 (func_801734BC ×134)

2026-07-12 21:51:27 -06:00

len_tells.py

feat(phase-31 T8): LEN+N lane — 587 near-misses routed; 345 wrong-drafts reclassified; detectors live

2026-08-14 19:54:57 -06:00

lenmiss_route.py

feat(phase-31 T8): LEN+N lane — 587 near-misses routed; 345 wrong-drafts reclassified; detectors live

2026-08-14 19:54:57 -06:00

lift_types.py

feat(phase-29): broad §20 type-lift lands — 154 types fleet-wide, R22 140/140 (§64)

2026-07-23 19:54:19 -06:00

lint_symbol_refs.py

fix(phase-30 S39): close the §134 class — the last two line-shape scanners route through cdecl._mask

2026-08-05 00:37:59 -06:00

lora_grind.py

perf(phase-23): lora_grind computes progress.py --fleet only on propagate sweeps (was every batch, ~14s overhead dominating small batches)

2026-06-30 15:32:56 -06:00

macro_draft.py

feat(phase-30 S47-B): conform 8 declaration axes (~10,930 sites); 3 guard defects fixed; 213/213

2026-08-10 16:01:49 -06:00

main_lane.py

fix(main-lane): tag namespace exhausted after 100 waves — extended with a second prefix (200 tags); the lane crash-looped for 2h while parked drafts waited (R51 finite-namespace class)

2026-08-26 11:05:02 -06:00

main_queue_rebuild.py

feat(tools): T5pre — main_queue_rebuild.py derives main's open set from progress.classify() (147 = 144 stubs + 3 NM; 959 LINKED + 2 blobs excluded) and attaches the newest draft per fn (136 queued, 11 never-drafted listed for the main wave; the old queue was 156/166 stale); campaign_status quotes the honest main denominator (P31 S62)

2026-08-26 15:02:15 -06:00

make_apicard_used.py

refactor(phase-9): T8a — de-default the EXE-curation helpers (thread --exe/--vram-base)

2026-06-15 20:55:28 -06:00

make_libgs.sh

feat(phase-7): Gen1 close-out — matching loop industrialized, PsyQ libs linked, LZSS matched (v1.7.0)

2026-06-15 13:44:32 -06:00

make_snd_used.py

refactor(phase-9): T8a — de-default the EXE-curation helpers (thread --exe/--vram-base)

2026-06-15 20:55:28 -06:00

masked_diff.py

fix(oracle): masked_diff kept the opcode at R_MIPS_26 slots — a masked j/jal was swallowing whatever the other side held

2026-08-29 12:56:20 -06:00

masked_scorer.py

feat(phase-24): T2 — floor-free relocation-masked permuter scorer (-drz)

2026-07-02 23:10:54 -06:00

match_one.py

feat(p31 s67): stranded-draft census + honest jtbl probe — the frontier's biggest class is carve plumbing

2026-08-31 01:46:16 -06:00

match_protos.py

feat(phase-3.5): prototype research spike — go/no-go (protos don't accelerate Gen1 matching)

2026-06-14 03:19:23 -06:00

mk_write.py

fix(tools): jr_isolate_all recognises lifted "} __attribute__((...)) X;" typedefs as engine types (5 drafts refused on Block4 after the lift) and drops a cut equal to the object start (a jr function that leads its object made an empty region -> out of order); mk_write refuses to merge a block that moved on disk since the caller read it (R43) (P31 S62 T3)

2026-08-26 15:48:50 -06:00

neighbor_ref.py

feat(cards): neighbor_ref.py — retrieve MATCHED functions as worked examples, ranked

2026-08-31 22:08:34 -06:00

new_binary.sh

feat(phase-30 S45 II.1a): all 38 MAIN modules onboarded byte-identical at the §S44 static addresses

2026-08-06 12:36:49 -06:00

new_overlay.sh

feat(phase-30 S44 I.1e): tools/new_binary.sh — one onboarder for every flat-blob class

2026-08-06 10:59:03 -06:00

normalize_self_decls.py

feat(phase-29): func_8013D53C 14/137 + NSD hardened (rewrite-to-canonical, (void) fix) + Cmd lift

2026-07-18 20:30:43 -06:00

o0_boundary.py

feat(o0): o0_boundary.py — the stranded-boundary -O0 sweep, and its honest null

2026-08-31 16:50:09 -06:00

o0_detect.py

feat(p31 s67): stranded-draft census + honest jtbl probe — the frontier's biggest class is carve plumbing

2026-08-31 01:46:16 -06:00

o0_subsplit.py

feat(phase-30): tools/o0_subsplit.py — the T2 carve-within-a-carve driver; +3 banked in ov_SC03_015

2026-07-31 08:41:54 -06:00

oracle_reorder.py

docs(phase-31): S53 journal harvest — §188 the epilogue tail is an ASSEMBLER artifact (corrects §177, answers §182), §189 five source-cited laws, §190 three prescriptions, §191 the rejected set; + tools/oracle_reorder.py

2026-08-17 09:55:46 -06:00

orchestrator.py

feat(phase-21): reach-1 harvest pivot — fuel built + handoff (×134 walls exhausted at 61.16%)

2026-06-22 21:22:35 -06:00

overlay_src_split.py

feat(o0): md_MAIN_003 carve — the module-binary -O0 route opens, func_800D0D6C banked (345 ins)

2026-08-31 18:13:42 -06:00

ox_campaign.py

fix(campaign): measure what the campaign can see — banked-today from the INCLUDE_ASM invariant, gate_min beside wall_min, a lock-aware fleet R22 window (P31 S61 T10+/S61-8)

2026-08-25 23:49:43 -06:00

p16_improve.py

feat(phase-16): permuter batch tester + known-answer improvement-loop workhorse

2026-06-18 23:40:17 -06:00

p16_known_answer.py

feat(phase-16): S5 driver+supervisor+safe-exit + NULL fix + sig_unify pipeline

2026-06-18 23:15:31 -06:00

p16_permute.py

fix(phase-30 S43): the §148 "GTE macro wall" is a SILENT CPP FALLBACK — permuter lane was dead on 63 drafts

2026-08-05 15:59:27 -06:00

pads_audit.py

feat(tools)+docs: T2 knowledge — interleave_check.py + pads_audit.py promoted (derive carve specs from the bytes; SETUP rows), cookbook §302 (the three carve-state invariants; five reds classified), frontier-s61 Addendum 3 (red list 5 -> 0, 39/40 held banks) (P31 S62)

2026-08-26 14:29:35 -06:00

parallel_gate.py

feat(tools): work_evidence — assert a tool ACTUALLY DID the work it reports

2026-09-01 18:59:04 -06:00

permuter_ils.py

feat(phase-24): T7 §G — 6/8 giants cracked, 4 ×134; fleet 65.75→65.91%

2026-07-07 20:20:43 -06:00

permuter_weights.py

feat(phase-29 Task-13B): func_80141B90 x138 + the reach repricing + the length profile

2026-07-21 15:05:56 -06:00

plumbing_groups.py

feat(phase-31 T6): PLUMBING campaign — +9 banked (64% probe); recipe + 3 laws (§173)

2026-08-14 19:42:55 -06:00

prefetch_fleet.py

fix(phase-30 S44 I.1d): the module class reaches every enumerating consumer

2026-08-06 10:57:17 -06:00

pregate_check.py

fix(phase-31): wire the S56 findings into the tools, SETUP.md and memory

2026-08-18 15:32:32 -06:00

progress.py

feat(phase-30 S45 L3-p3): SC02/9 = the Steam Knight boss module — decoded, captured, retro-verified, onboarded; parked = 5

2026-08-07 15:07:55 -06:00

psyq_build_libs.sh

feat(phase-7): PsyQ library-linking proven byte-exact + rodata-island mechanism — session C checkpoint

2026-06-14 21:24:10 -06:00

psyq_identify.py

refactor(phase-9): T8b — remove transitional defaults across link pipeline (no-EXE-default end state)

2026-06-15 21:00:36 -06:00

psyq_integrate.py

fix(psyq_integrate): main was RED on every incremental relink — make the externals file monotonic

2026-08-31 17:03:50 -06:00

psyq_lib_split.py

feat(phase-7): PsyQ library-linking proven byte-exact + rodata-island mechanism — session C checkpoint

2026-06-14 21:24:10 -06:00

psyq_link_lib.py

refactor(phase-9): T8b — remove transitional defaults across link pipeline (no-EXE-default end state)

2026-06-15 21:00:36 -06:00

psyq_link_region.py

refactor(phase-9): T8b — remove transitional defaults across link pipeline (no-EXE-default end state)

2026-06-15 21:00:36 -06:00

psyq_link.py

refactor(phase-9): T8b — remove transitional defaults across link pipeline (no-EXE-default end state)

2026-06-15 21:00:36 -06:00

r22_verify.sh

fix(r22 guard): record liveness, do not infer it — my mtime heuristic failed BOTH ways

2026-08-31 18:59:50 -06:00

ram_probe.py

feat(phase-17): T4+T5 — actor struct recovered (comprehension) but byte-neutral for matching

2026-06-19 13:03:14 -06:00

reconcile_slate.py

feat(phase-31): typedef comparison ignores comments (gate_main + reconcile_slate); func_8001ABBC + func_80037028 banked via byte-neutral TU edits

2026-08-16 22:49:05 -06:00

reconcile_tu.py

fix(phase-30 S1e): the distinct-code "regression" was a STALE DIGEST — alias lever ungated

2026-08-04 21:48:57 -06:00

recover_drafts.py

feat(phase-30 S48): tools/recover_drafts.py — the transcript IS the backup

2026-08-11 21:12:44 -06:00

recover_giant.py

docs(phase-22): T3 — distill §28 canonical-extern recovery + recover_giant.py + decomp.wiki patterns

2026-06-26 03:06:54 -06:00

recover_integration.py

feat(integration): teach the rest of the toolkit about §378 (self-caller cast)

2026-09-01 11:03:06 -06:00

recover_rejects.py

fix(recovery): validate the binary before staging; aprop_autodraft adopts the TU's spelling and refuses asm bodies

2026-08-24 21:33:32 -06:00

reloc_identity.py

fix(aprop): reloc --fix keyed by (binary,fn); lane gains rtu second-chance, MISMATCH auto-repair, borrow-tier decls

2026-08-24 22:29:29 -06:00

reloc_verify.py

feat(phase-29): promote tools/reloc_verify.py — resolve every relocation before paying a gate cycle

2026-07-27 00:06:01 -06:00

residual_class.py

fix(phase-29): T54 — correct the ADDRESSING route, and fix the reason changing it was inert

2026-07-28 19:34:00 -06:00

residual_rules_b.py

feat(integration): teach the rest of the toolkit about §378 (self-caller cast)

2026-09-01 11:03:06 -06:00

residual_rules.py

feat(triage): the residual-classifier head-to-head — both implementations, kept

2026-08-31 22:24:56 -06:00

restore_dropped_decls.py

perf: -j on the last three per-binary make build sites (family_sweep/twin_sweep, rollout_o0, restore_dropped_decls)

2026-08-31 14:24:03 -06:00

rollout_o0.py

perf: -j on the last three per-binary make build sites (family_sweep/twin_sweep, rollout_o0, restore_dropped_decls)

2026-08-31 14:24:03 -06:00

rtu_match.py

fix(rtu_match/blocker_probe): main was structurally unprobeable — pass the TU path and asm subdir, never reconstruct them

2026-08-31 16:34:01 -06:00

rtu_second_chance.py

fix(aprop): S59 lane overhaul — four measured defects, each byte-diagnosed before the fix

2026-08-24 22:22:12 -06:00

rtu_shadow.py

fix(gate/tools): baseline-RED skip in the gate, jtbl_pads_fix un-broken three ways, mk_write parse guard, shadow join for 0-bank waves (P31 S61 T10+/S61-10)

2026-08-26 00:29:05 -06:00

scan_leftovers.py

feat(phase-31): tools/scan_leftovers.py — re-verify unbanked drafts (found 37 free MATCHes / 3,240 ins); cookbook §180

2026-08-16 18:11:32 -06:00

scope_data_externs.py

fix(phase-30 S48): scope_data_externs spliced carried externs INTO a comment

2026-08-11 19:07:07 -06:00

scope_demote_drafts.py

feat(p31 s67): stranded-draft census + honest jtbl probe — the frontier's biggest class is carve plumbing

2026-08-31 01:46:16 -06:00

scope_tu_externs.py

feat(phase-29): T56 — wire the tu-scope lever into family_sweep; func_80144090 0/136 -> 136/136

2026-07-28 20:15:11 -06:00

seed_ref.py

feat(seed_ref): the CONTAINED tier + docs for the twin ladder (§390/§391, accelerator #18)

2026-09-01 13:06:27 -06:00

serve_local.py

feat(phase-23): local GPU serving + prompt-fix + corpus-v3 (macro+struct mining)

2026-06-30 01:46:54 -06:00

shared_lock.py

feat(phase-30 S45p7): Stage 1 complete — shared-state RW lock + 15 more banks (wave-3 revived)

2026-08-07 18:57:10 -06:00

sig_image.py

feat(phase-31 T3): main enablement — sig-main at splat-true lengths + main streams

2026-08-14 17:48:51 -06:00

sig_unify.py

fix(phase-26a): A3e — gate_stage pinned the byte-gate back to 4.9%, OF A3'S OWN FIX

2026-07-14 15:39:07 -06:00

split_src_region.py

refactor(phase-9): T8a — de-default the EXE-curation helpers (thread --exe/--vram-base)

2026-06-15 20:55:28 -06:00

strand_census.py

feat(p31 s67): stranded-draft census + honest jtbl probe — the frontier's biggest class is carve plumbing

2026-08-31 01:46:16 -06:00

stub_invariant_audit.py

feat(tools)+docs(phase-31): T1 — resolver declfix wiring; masked_diff/rtu_match compare internal-j targets (jrel; §195 premise refuted, cookbook §301); diff_autopsy.sh + stub_invariant_audit.py promoted; frontier-s61 Addendum 2 (step-1 falsifier fired: 54 carve-refused, 2 DIFF autopsied+banked, 1 plumbing); S62 log

2026-08-26 14:04:54 -06:00

sweep_citations.py

fix(phase-29): T36/T37 — my checker had a FORM-FEED bug; the "12 fabricated" were mine, not the agents'

2026-07-28 13:20:27 -06:00

sweep_parallel.py

perf(gate): the parallel gate was serialized by its own shared lock — -j was decorative

2026-08-25 10:43:10 -06:00

symcheck.py

feat(phase-29): tools/symcheck.py — the pre-gate symbol-set guard (§67a)

2026-07-24 21:30:08 -06:00

t5_bank.sh

feat(tools): fix_decl_mirror.py — the MIRROR of fix_tu_ret_decls (TU declares a VALUE return for a genuinely void definition; it SKIPped in every S63 wave). Narrowing the TU decl is NOT byte-neutral there (a caller reads $v0), so the repair is draft-side: define the body under a distinct identifier bound by __asm__("func_X") — the tree's own impl_* idiom. T0 blast radius. R39 negative control found and killed TWO real defects (block-scope decls: 856 false positives; sibling-TU decls: 580) and now passes 78,727/78,727 clean. Wired into t5_bank.sh (P31 S64)

2026-08-27 12:19:36 -06:00

t5_cards.py

docs(p31 s67): wave-playbook (the CURRENT pipeline) + seed_ref fix + harvest §333-§338

2026-08-31 13:48:56 -06:00

t5_distill_args.py

fix(distill): the novelty selector was inverted, and it could not see the notes worth reading

2026-08-29 15:22:14 -06:00

t5_targets.py

fix(t5_targets): dedupe residues by (binary,fn) across prior waves and enforce one fn NAME per wave across residues too — the t5i draw produced 4 duplicate names (2 the same fn from two waves) and the name-keyed pack builder correctly REFUSED (R43/R48); deferred duplicates are reported, not dropped (P31 S63)

2026-08-27 10:25:01 -06:00

t7_bank.py

feat(phase-25): T7-M1 — 37 exemplars banked ×1 via canon_sig_reconcile v3.2 (giants incl. func_8016A290/284)

2026-07-10 00:02:55 -06:00

test_jtbl_parse_config.py

fix(jtbl_carve): parse_config now implements its documented contract (md_* c-line corruption)

2026-08-24 10:42:11 -06:00

test_o0_detect.py

feat(o0): md_MAIN_011 compiles -O0 — 21 functions / 4,321 ins unlocked, byte-proven

2026-08-24 12:43:40 -06:00

test_reconcile_ledger.py

test(phase-30 S45p7): prove the reconcile-ledger undo — the owed negative control

2026-08-07 18:50:29 -06:00

test_residual_class.py

feat(phase-29 Task-13B): func_80141B90 x138 + the reach repricing + the length profile

2026-07-21 15:05:56 -06:00

train_lora.py

feat(phase-22): train_lora auto-cleans GGUF intermediates (the ~30GB that crashed C:)

2026-06-29 20:07:48 -06:00

treelock.sh

fix(phase-30): treelock.sh — an flock MUTEX for tree-writing campaigns (incident 2: a poll is not a mutex)

2026-07-30 22:00:32 -06:00

triage_ladder.py

feat(triage): the triage ladder — built, wired, and acceptance-green

2026-08-31 23:53:50 -06:00

twin_rescan.py

feat(tools): verify_binary + twin_rescan — put S69's two habits in the tooling, not in prose

2026-09-01 14:50:26 -06:00

twin_sweep.py

feat(twin_sweep): refusal ledger — stop re-gating candidates the gate already refused

2026-08-29 19:14:02 -06:00

uniquify_type.py

feat(phase-29): VARIANT camps -> UNIQUIFY (not reconcile); validated on Buf, R22 140/140 (§64a)

2026-07-23 20:51:42 -06:00

validate_targets.py

docs+rules(S58): R42 commit-banked-work-immediately, R43 refuse-unsupported-input

2026-08-23 12:59:59 -06:00

verify_binary.py

feat(tools): verify_binary + twin_rescan — put S69's two habits in the tooling, not in prose

2026-09-01 14:50:26 -06:00

verify_map_findings.py

fix(phase-29): T36/T37 — my checker had a FORM-FEED bug; the "12 fabricated" were mine, not the agents'

2026-07-28 13:20:27 -06:00

verify_worktree.py

docs+rules(S58): R42 commit-banked-work-immediately, R43 refuse-unsupported-input

2026-08-23 12:59:59 -06:00

wall_sweep.py

feat(walls): wall_sweep.py — ENUMERATE the §332 delay-slot macro walls, 10 fns / 1,027 ins

2026-08-31 19:59:27 -06:00

wall_taxonomy.py

feat(phase-17): T3 — bank 42 sig_unify matches + propagate x134 (fleet 54.48%->55.00%); fix taxonomy classifier

2026-06-19 12:13:38 -06:00

warmstart.py

feat(phase-31 T9): warmstart feeder + weak-cards — the campaign lanes are armed

2026-08-14 19:59:21 -06:00

wave_args.py

feat(triage): the triage ladder — built, wired, and acceptance-green

2026-08-31 23:53:50 -06:00

wave_card_fuel.py

fix(wave): cards keyed by (binary,fn) — t5a received 48/48 ANOTHER binary's card (R48); t5_cards.py builds the target's OWN fuel (tu_ref 88% / decl_prior 98%) via wave_card_fuel.py extracted from build_wave_atlas (R33); t5_bank.sh recovers un-banked drafts via fix_tu_ret_decls (2/2 on t5a); distill workflow generalized (P31 S63 T5)

2026-08-26 20:19:41 -06:00

wave_judge.py

feat(tools): T5 wave draw/bank halves — t5_targets.py (routed, ledgered, band-stratified draw) + t5_bank.sh (judge → R22 sweep → commit); claude_wave_draft.js generalized; packs/judge refuse duplicate fn names (R43/R48); sweep #8 213/213 (P31 S63 T5.1)

2026-08-26 19:35:54 -06:00

wave_snapshot.py

feat(phase-31): wave Z closed — 127 banked, R22 213/213, fleet 96.1/91.8/97.11

2026-08-18 15:25:49 -06:00

wave_targets.py

fix(phase-26a): A3 — target selection was blind to 91.6% of the remaining work; now derived

2026-07-14 09:22:31 -06:00

work_evidence.py

feat(tools): work_evidence — assert a tool ACTUALLY DID the work it reports

2026-09-01 18:59:04 -06:00

worklist.py

fix(phase-28 T7): worklist --assert-partition read the wrong manifest key (silent default)

2026-07-16 02:01:15 -06:00