c0e8fbdaae
phase-36: T6 — the rung-D instrument fixed (the target was assembled from a listing), rung R built (R2/R3/R4/R5)
...
THE INSTRUMENT (R40). Two campaigns returned "0 of 16" with a straight face; both were the harness. The target object had
been assembled from a DISASSEMBLY LISTING, which is a second toolchain with its own answers:
- objdump prints the pseudo-instruction `move` for `addu rX,rY,$zero`; gas assembles `move` as `or` — 24 wrong words in
one 234-instruction function, silently;
- a listing's %hi/%lo pairs come back RESOLVED with no relocation, while every candidate carries one, and the masked
scorer compares reloc operands.
The permuter therefore scored 28 for a body that IS byte-identical: score 0 was unreachable and every NO-MATCH was its own.
- tools/delever_permute.py: the target is now the tree's OWN (levered) body compiled by the build's tail into a
one-function object — the candidates' relocations by construction — and `match_one` must call that body a MATCH against
the ROM listing before the search starts (R34 keeps it from being circular). Base score for the tree's own body: 0.
- tools/p16_permute.py `setup(target_o=)` + tools/permuter_ils.py `--target-o` (defaults unchanged).
- `--positive-control TU FN`: perturb a matching body by one commutative swap, require the permuter back to 0.
- tools/verbatim_target_s.py --gas now VERIFIES itself: assemble, disassemble, compare word by word with the image,
`.word 0x…`-patch what does not reproduce (24 in that function), REFUSE what still disagrees. The listing is a public
artifact (decomp.me) and was wrong for every function containing a `move`.
THE PROFILE. The weight profile now comes from the register a needed pin names, not just the site kind: callee-saved
($16-$23) is an allocation-order residual -> regalloc; caller-saved ($2/$3/$4-$7) is not -> cse. Read from the bytes: the
residual on func_80163EC8 (`register … __asm__("$2")`) is `and v0,v1,v0` against `and v0,v0,v1` — the operand order of one
`&` — and the regalloc profile weights perm_commutative 2.0 while cse weights it 40.0.
RUNG R (tools/delever.py --recipes): the cookbook's byte-neutral shape recipes, mechanically, seeded with the body's
lever-free text — R2 the formerly-pinned declarations permuted, R4 one moved through the whole declaration run, R3 an
initializer split placed after the run (C89), R5 the operand order of one commutative operator (the caller-saved lever, and
the only recipe needing no pinned declaration). Identity control on both the splice and the oracle before any verdict;
markers scrubbed within the banked body's own span only; selftest cases on a fixture whose answers are known by hand.
Also: a threading race in the site cache published the empty dict before filling it and made a whole batch report
"no site in this TU"; the file-scope asm dropper took an asm-LABEL clause for a statement. SETUP + dictionary rows.
2026-09-09 11:08:20 -06:00
6e1677b4da
phase-36: T6 rung D built and calibrated — tools/delever_permute.py, the GTE levers on the ladder (462/462), four harness defects
...
- tools/delever_permute.py: one exemplar per RESIDUE text class from the ledger (copies desc, needed asc) prepared as a
single-function TU (delever's rung-A rewrite; other definitions -> prototypes; shared-header includes -> their prototypes;
INCLUDE_ASM and file-scope asm dropped; the build's own CPPFLAGS through cpp -P), the target regenerated from the ROM image in
BOTH forms (--gas for target.o, splat for match_one), permuter_ils with the profile from the NEEDED kinds, a winner banked only
through delever --apply-body + the GTE re-fold. Scratch/winners keyed alias+fn (R48).
- the control (R39/R56, new): every attempt first requires the LEVERED body to be match_one MATCH against the regenerated target,
then records the lever-free body's distance. --calibrate --limit 12: 12 of 12 MATCH; starting distance min 8 / median 78 / max 276.
- delever: ("B","gte-lever") joins REMOVABLE — a direct statement's clobbers reset to its canonical set, a variant-macro use pointed
at the canonical macro whose name comes from the variant definition's SIGNATURE (gte_rt_m -> gte_rtv0tr, not Sony's gte_rt);
462 of 462 gte-lever sites now offered to the ladder, 0 before. gte_consolidate.canonical_match() is the one reader of the
canonical table (R33), direct_rewrite refactored onto it; both selftests green.
- p16_permute.setup(outdir=) + permuter_ils --pd: a scratch dir keyed by the caller, defaults unchanged.
- four harness defects found by running it: the splat listing is not assemblable (R98 in a second place); pycparser rejects
__attribute__ and the permuter then silently permutes nothing; include_asm.h injects a file-scope .include "labels.inc" that
collides with the permuter's own macro.inc; an asm-LABEL clause is not an asm statement (a bare scan ate one and left a headless
K&R body). SETUP row (R21), dictionary row (R87), .gitignore allowlist for the outcomes ledger.
2026-09-09 10:21:16 -06:00
d025e67e71
phase-36: T5 CLOSE — the GTE consolidation and the dead-macro sweep: include/gte_inline.h (50 canonical macros for 9,102 definitions), 8,951 per-TU definitions deleted, 629 clobber variants freed byte-identical and 70 kept as marked levers, 575 direct statements → Sony-named calls, 274 dead launder macros swept; the census learns cross-file macro names (10 launders surfaced in two shared headers, 6 removed); per-TU asm macro definitions 9,540 → 314 with 0 canonical duplicates; THE NUMBER 34,091 sites (incl. 462 GTE levers) in 12,712 bodies, all marked, 0 orphans — lever_census --check OK; R22 218/218; SETUP rows; kit corpus; the 🛑 block for T6
2026-09-09 07:12:06 -06:00
3b2c5178b6
phase-36: T5 — the scrub of 3 markers orphaned by the rejudge (R22 218/218, r22_sweep1.log); delever --scrub --dirty-ok; the sweep's dead-macro test skips a name's sibling definitions (SHB is defined twice per unit, so nothing was ever dead); kit corpus
2026-09-09 06:38:42 -06:00
5b6e1a2fe7
phase-36: T5 — the consolidation applied (644 files: 8,951 per-TU GTE definitions gone, 629 clobber variants byte-identical without the clobber, 70 kept as marked levers, 575 direct statements → canonical calls, 85 header-bound homonyms kept) + the marker repair (scrub 344 misplaced, re-mark 349); four instrument fixes: scoped renames, header-bound homonyms kept, object-like macros, the canonical table stable (the header's definitions in the inventory) and the census cache keyed on the tables; R22 218/218; the number 34,090 (33,625 + 465 GTE levers) all marked, 0 orphans, --check OK; kit corpus; log entry
2026-09-09 06:29:00 -06:00
0bd784c64e
phase-36: T5 (header step) — the GTE consolidation: tools/gte_consolidate.py (signatures by the build's own maspsx→as tail: bytes + operand counts + clobbers; one canonical text per signature under Sony's names, Sony's clobbers canonical even when every definition carries the steer; lever variants <name>_m tried as canonical first; direct statements → canonical calls; --sweep), include/gte_inline.h (50 macros for 9,102 definitions) included from common.h — R22 218/218; the census's gte-lever class + per-TU definition count + strict gate; the cycle's MODE=gte; dictionary + SETUP rows; kit corpus; log entry
2026-09-09 05:28:50 -06:00
917c68afd9
phase-36: T4 CLOSE — the mechanical campaign over the whole population: 16,334 bodies judged in 13 gated batches (3,341 lever-free = 20.5 %; 21,063 of 56,445 sites removed or rewritten = 37.3 %; 8,239 replays, 1 disagreement; 49,487 compiles); the file-scope asm statements judged too (batch tus11: 6 barriers NEEDED, 7 .section blocks refused as asm-data and marked); THE NUMBER 53,234 → 33,625 sites (−36.8 %) in 12,501 bodies (1,728 distinct), 33,625 marked, 0 UNMARKED, 0 orphans — lever_census --check OK; the residue 12,967 bodies in 1,803 distinct classes (pins-only 6,018 · pins+asm 4,310 · asm-only 2,141 · C/D-only 498); published by make report (README line 30, progress.json counts.levers, the timeline); kit corpus; the 🛑 block rewritten for T5
2026-09-09 05:05:33 -06:00
fa008b8629
phase-36: T4 — the census's known-true controls keyed on their bodies' T1 text hashes (N-A once the campaign edits a control body: tus7 removed func_80184034's three bare-name pins and the census refused a correct tree); the cycle's FINISH=<label> mode completes a batch from its green logs (tus7 finished so: 34f03003d); kit corpus; log entry
2026-09-09 04:04:14 -06:00
77a1298cb7
phase-36: T4 — batch tus3 killed mid-apply by the harness's low-memory guard (22 GB free): recovered by dropping its 641 ledger rows (backup) and --restore (50 files, src clean); --restore now drops the in-flight label's rows itself (inflight.json carries the label); the cycle's usage: run it DETACHED (setsid nohup) with a tiny waiter; SETUP row; kit corpus; log entry
2026-09-09 03:35:35 -06:00
dd01cbdf6e
phase-36: T4 (tool fixes between batches) — 'done' is per body (tu, fn, after-hash): keying it by text alone let every fleet-wide copy of an all-NEEDED exemplar pass as done unmarked (2204 → 2483 drawable files); the includer map cached per file (29 s → 1.3 s per batch, equal to the uncached map); the cycle waits out a transient .git/index.lock and clears a stale one (batch tus2's commit was lost to one and committed by hand, 85905840c); kit corpus; the log entry with tus1/tus2 — the number 46,424 sites in 14,758 bodies · marked 9,393 · unmarked 37,031
2026-09-09 03:30:53 -06:00
e5f59c12f4
phase-36: T3 close — the campaign tool proven end to end on ov_SC04_011 (batches t3_tus1 + t3_tus2, R22 218/218 each): the zero-register use composed into its host edit (the one refused body redrawn via --redraw), a removed site consumes its trailing !FAKE marker, the census counts ORPHAN markers (--check fails on them; the walk cache keyed on the tool's hash — it had hidden them) and delever --scrub removes them (2 scrubbed here, object identical, R22 218/218 wall 143 s); the cycle's greps fixed (no ^ anchor, an empty verify line refuses, THE PHASE'S NUMBER line, entries before the checkpoint block); the number 53,033 sites / 15,638 bodies (2,246 distinct) · 310 marked · 0 orphans; SETUP + dictionary rows; kit corpus; the t3_tus1 entry rewritten from the logs (R66); the 🛑 block rewritten for T4
2026-09-09 03:14:30 -06:00
39e3e1851f
phase-36: T3 (tools) — tools/delever.py the campaign tool: --plan/--apply (the file is the write unit, its final compile through every recipe the proof; 'done' = a ledger fact keyed by the body's normalized text before/after; replay from an exemplar in one compile; !FAKE markers on class A/B survivors; --restore from inflight.json; --apply-body for T6/T7; a selftest with a stub oracle), the launder-ASSIGNMENT rewrite (989 of 4,793 direct launders — deleting them made cc1 2.7.2 abort: the probe's 'Done' error, reproduced 3/3), macro shapes (SHB deleted, a launder expression valued, compound macros refused for T5), the oracle's COMPILE-CRASH verdict + judge_all + the mtime restore, the census's src_stamp/head, tools/delever_cycle.sh (apply → R22 → census → log + headline → commit); selftest OK; census 53,234 unchanged; calibration 177/177; plan ov_SC04_011 = 18 files / 176 bodies / 540 sites; SETUP + dictionary rows; kit corpus; T7 waves need Drew's direct approval (decisions)
2026-09-09 02:45:58 -06:00
617f29e30c
phase-36: T2 — the oracle and the probe: tools/delever_oracle.py (4,284 recipes via make -n -W; a candidate compiled in place to a scratch object and compared with the fleet run's; calibrated 177/177 untouched + twin 32/32 + positive control DIFFERS) and tools/delever.py (positional rewrites per lever class, rung A strip-all + rung B greedy, headers serial after the TU phase); --probe on 283 distinct bodies: A identical 53/283, sites removed 561/1,309 (42.9 %), needed 748 (launder 77 %, barrier 82 %, pins 45 % — v0/v1 + a0–a3 dominant), 0.225 s per compile — T4 priced at ~94k compiles ≈ 0.5 h, the residue ≈ 81 % of bodies ≈ 1,800 distinct exemplars; dictionary + SETUP rows; kit corpus; the 🛑 block for T3
2026-09-09 02:03:04 -06:00
a1407bdfc1
phase-36: T1b — the verbatim manifest learns the in-function form (49 rows: 27 PERMANENT hand asm incl. 22 per-overlay scratchpad stack-switch trampolines, 8 Sony GsTMDfast* routines carried as asm, 11 DECOMPILE-NOW + 3 UNCERTAIN game bodies never decompiled); verbatim_check detects the form structurally via lever_census.whole_body_asm_functions (--strict: 2,709 bodies, no drift); the census defers exclusion to the manifest's dispositions — the phase's number 53,234 sites in 15,679 bodies (2,227 distinct); progress.py counts PERMANENT (27) + SDK (8) with a dated correction; README/wiki/verification/gen3 docs corrected (14 game functions were asm in C shells); kit corpus
2026-09-09 01:44:31 -06:00
357f8a1ed3
phase-36: T1 — tools/lever_census.py, the self-asserting lever census (coverage asm 80,396 = 66,415 live + 7,189 macro-block + 6,792 comment/dead; 4/4 controls; the five manifest bodies excluded): 53,221 pin/asm sites in 15,666 bodies (2,214 distinct) — 37,720 register pins + 15,501 asm statements, 0 marked; 6,911 GTE ops and 23 whole-body asm routines in C shells (2,683 sites, outside the manifest — Drew's call) set apart; progress.py levers block + README sentence; dictionary + SETUP rows; kit corpus; the 🛑 block for T2
2026-09-09 00:52:22 -06:00
ac8c29edb1
src(phase-35): T5b — the text tier: 38 same-address functions whose bytes vary per binary (the TU's declaration environment carries the original per-overlay variance) share one text each through h_text registry groups with per-member h_exact — 1,957 private copies in 1,540 TUs replaced by the include, 38 headers (func_<VRAM>__t<hash8>.h), 138/138 binaries green per binary, 0 rejected; dedup_integrate C1 per member (3173 validated, 0 failed); share_census indexes h_text groups by site, exports text classes, the S1 text half is a violation (0 now), the census json carries the oracle's counts; share_body --bucket text; progress.py/README: 103,015 bodies written once, 38 text-tier functions / 2,030 sites, 1,668 copies of 381 tiny bodies deferred; R22 clean fleet 218/218 in 90 s; SETUP; kit corpus
2026-09-08 22:39:42 -06:00
c8224e657f
tools(phase-35): T7 — the S1 invariant wired into tools-health (share_census --selftest + --check --strict-macros --strict-text, by exit code), C2c/C2d in dedup_integrate --check (one source under src/shared/ defining one function; every member's site includes it — from the census's per-instance forms), progress.py: unique_function_bodies 105,007 / duplicate_source_copies 160 in 51 ledgered classes + the dated corrections list + the README sentence; the second oracle refined (distinct TUs; deferred vs pending vs violation): 380 texts deferred inside cross-address classes, 38 same-address texts / 2,030 sites byte-variant per binary (PENDING the owner), 0 violations; negative control in place: one reverted include -> S1 FAIL naming the class + C2d naming the member; tools-health OK (551 s)
2026-09-08 21:46:05 -06:00
db63fcc673
tools(phase-35): T6 part 2 — the dead macro branches dropped from 12 live tools; the macro-form guard in tool_census --check (ast scan of every LIVE tool, FROZEN/retired/6 whitelisted detectors excluded; 0 LIVE tools reference the retired form; --guard-root negative control: the pre-T4 tree flags exactly the 12 fixed + 3 retired); progress.py: an empty-bodied included member is EMPTY, not folded into REAL (the R32 bucket assertion caught the double count) — the fleet denominator 363,221 -> 363,680 (+459: SC03_015's 219 single-site macros, SC03_118 +3, the never-extended members of five groups) and REAL 360,744 -> 350,533 with EMPTY up the same; progress.json + README block + fleet table regenerated; SETUP; kit corpus
2026-09-08 20:53:05 -06:00
5a9f455e60
tools(phase-35): T6 part 1 — 14 macro-era tools FROZEN (tools/frozen.py: one refusal from main(), imports never exit; the plan's 7 + 7 the guard census found, family_remap's CLI among them), 4 retired to tools/sunset/ (dedup_propagate + dedup_extend -> share_body.py with add_members_surgical and the load_sig library moved verbatim; macro_draft; test_reconcile_ledger), six lane callers repointed to share_body --apply --bucket new, validate_targets imports share_body, tool_census accepts FROZEN; dictionary + sunset README rows; kit corpus
2026-09-08 20:34:41 -06:00
cd43347791
config(phase-35): T5 CLOSE — the registry behind the source repaired (639 including-but-unlisted members added to 5 groups, the 5 verbose groups converted to shorthand; add_members_surgical refuses a verbose group, R43); S1: 10,180 classes, 10,180 satisfied, 0 violations (share_census --check exit 0); same-vram backlog 1,099 -> 51 classes = the ledger (50 TU-CONFLICT + 1 GATE-REJECT); dedup-check 3135/0; audit-binaries OK; registry 2,220 -> 3,135 groups; the tail run (bucket new empty); cycle logs; SETUP (the context guard); kit corpus
2026-09-08 20:24:28 -06:00
e4b25e5f70
tools(phase-35): T5 — share_body_cycle.sh, the unattended batch cycle for bucket new (batch → exit code + gated N/N → phase-log entry → commit → clean fleet run every N batches, stops on the first red); share_body --label; dictionary row + SETUP; kit corpus
2026-09-08 19:53:09 -06:00
a60340c6d4
src(phase-35): T5 — E_func_80168B70 re-exemplared from the majority text (share_body.py --reexemplar: header ← ov_SC01_000's copy, 7/7 includers green first, then 134 private copies → the include in 129 TUs, 141/141 binaries green, 0 rejected; the minority-header ledger row dropped); bucket 0 closed: 183 classes → 135 fully shared, 48 ledgered TU-CONFLICT (late-overlay declaration conflicts, for the types phase); SETUP paragraph (--batches, snapshot restore, the cause rule, --repair-registry, --reexemplar); kit corpus
2026-09-08 19:41:40 -06:00
5325fd506b
src(phase-35): T5 bucket 0 run 3 — 78 private copies in 24 TUs replaced by the shared include, 87 members extended (141/141 binaries green per binary); share_body: a twin's instance queued the same (tu, line) edit twice and the second replacement swallowed the next function (7 classes rejected in the five twin primaries on that artifact; now deduped), chained file:line: warning lines never a cause; the ledger re-judged by the fixed tool: 49 rows with real diagnostics (conflicting types / too few arguments in the late overlays), E_func_80168B70 kept with its cause pending the exemplar decision; kit corpus
2026-09-08 19:31:19 -06:00
090f8fb428
config(phase-35): T5 — the registry repaired: 378 listed-but-private members removed from 79 groups (share_body.py --repair-registry, derived from the census on (binary, vram); the 317 ledgered rejected pairs + 61 sites the S94 bisects wiped); dedup-check 2221/0, C1 256,120/256,120; census A 1,812 · B 183, same-vram backlog 1,099 -> 999 classes; the first cut's name-only match (9,269 removed, 14 groups emptied) caught by dedup-check + a known-true control and reverted before commit
2026-09-08 19:20:48 -06:00
f4163518a2
tools(phase-35): T5 — share_body.py: the failure cause is the first non-warning file:line diagnostic (gcc 2.7.2 errors carry no 'error' token — 254 of 303 rejections had read 'Error 33'), reason codes SYMBOL-NAME / GATE-REJECT / PARSE-ERROR / TU-CONFLICT derived from it and recorded with the cause in every ledger note, the bisect restores from an in-memory pre-edit snapshot (never git checkout — it wiped the previous batch's uncommitted shares), one batch per invocation by default; controls on the real red cc1.err + six synthetic logs; ledger header + kit corpus
2026-09-08 19:14:45 -06:00
571374b979
src(phase-35): T5 bucket 0 (first pass) — 788 members extended into their registry groups, private copies replaced by the include where the per-binary gate held (141/141 and 146/146 binaries green after bisect); share_body: the bisect now re-applies the kept classes in one bottom-up pass per TU (its per-class sequential re-application had shifted later classes' lines and rejected 125 of 183 classes on its own artifacts), the failure detail from error lines only, rejected extend classes ledgered, groups extended only with members that passed
2026-09-08 18:30:29 -06:00
0bccef9016
tools(phase-35): T5 — share_body.py, the permanent share tool for the include-at-site form (exemplar by majority text, header written once, private copies → include at their position, per-binary gate with object comparison and bisect, registry appended by text, exception ledger); the R37 probe: func_801681FC shared by ov_SC04_008 + ov_SC05_009 (2/2 binaries green, group S_func_801681FC); the census exports per-instance line ranges; bind_alias_header factored out of macro_to_header; SETUP + dictionary rows; kit corpus
2026-09-08 18:12:41 -06:00
a8457663fc
src(phase-35): T4 finalize — engine_core.h, ov_setters.h and clearTbl40.h deleted; the legacy sites converted (clearTbl40 as the SHARED_FN parameterized control in src/800_c.c; the three SC01_005 accessors as headers), the whale header moved under ov/ with every -O0 includer rewritten, 7 alias-form bodies given their own __asm__ binding, the registry's source/func lines text-edited by id, three non-shared headers rewritten; --verify OK (0 macro sites, no macro header); R22 clean fleet check-all: 218 passed, 0 failed of 218 in 145 s; dedup-check 2220/0; audit-binaries OK; the census 362,389/362,389 with 0 macro sites; SETUP + dictionary rows; kit corpus
2026-09-08 18:06:04 -06:00
c53a9e1a91
src(phase-35): T4 — every DEFINE_func_X() macro body becomes a plain-C header under src/shared/<space>/ included at its site (sotn's shape): 2,215 headers (ov/ 2,204, three module slots 12; 86 with the stable h_exact suffix), 246,347 sites in 3,818 TUs, the engine_prelude.h include in place of engine_core.h, 1,300 dead macros dropped; L0/L1: 213 binaries BYTE-IDENTICAL, 4,121/4,121 objects byte-identical before and after (tools/macro_to_header.py --apply, .run/P35/convert/ledger.txt); engine_core.h and the legacy headers still present until --finalize
2026-09-08 17:51:39 -06:00
2648aa5a92
build(phase-35): T3 — ov_SC04_019 TWIN_OF ov_SC04_018 with the primary's carve: yaml, JTBL_INTERLEAVE and the per-object JTBL_PADS lines regenerated from ov_SC04_018 by alias substitution (the twin's retired jump-table splits with them), create_c_files: False, src/ov_SC04_019/ deleted; twin + primary BYTE-IDENTICAL; interleave_check ALIGNED + pads_audit on the twin (R60; pads_audit now reads a twin's source through the oracle); dedup-check 0 failed; audit-binaries OK
2026-09-08 17:21:44 -06:00
9b0816e74a
build(phase-35): T3 probe — ov_SC01_006 is a TWIN of ov_SC01_005: one source directory per payload. Makefile twin rules (build/src/<twin>/<twin>%.o <- src/<primary>/<primary>%.c, same recipe, -O0 objects kept), overlays.mk TWIN_OF + SRC_DIR, the twin's yaml create_c_files: False, src/ov_SC01_006/ (30 files) deleted; twin + primary BYTE-IDENTICAL (56760dbe), the concurrent race test green, every consumer sees the twin through the oracle; share_census: copies = distinct sites, TWIN-COVERED reported (575 classes for the pair)
2026-09-08 17:10:00 -06:00
6cb056c932
tools(phase-35): T2 — the health chain learns the include-at-site and twin forms while the macro form still builds: one source-dir oracle (corpus.src_dirs/src_dir/twin_of from the Makefile), share_census.header_defs as the one reader of the include form, progress.classify counts include sites (fold keeps 255,632), overlay_src_split's include item kind + loud macro-table refusal, dedup_integrate C2a' (a header must DEFINE func), audit_binaries prelude + twin citizenship (CHECK 3b), cdecl differential not-applicable when the macro header is gone, recursive shared-header globs; the count correction 363,214 -> 363,221 (7 whale include sites never registered; ov_MAIN_012 2,401 -> 2,402) regenerated into progress.json + README
2026-09-08 16:46:55 -06:00
1dbffee87d
tools(phase-35): T1 — share_census.py, the fleet census of byte-identical function classes + the S1 'one source per unique function' checker with its exception ledger; 362,389/362,389 sig instances covered, 10,180 classes (A 1,712 · B 282 · C 6,107 · D 1,861 · M 218), self-test 7/7, four controls as measured; SETUP + dictionary rows, kit corpus regenerated
2026-09-08 16:24:08 -06:00
cd768d717a
tools(phase-34): cdecl --audit — create its scratch dir and refuse (never misgrade) when the gcc adjudication cannot run; the prune had removed .run/audit and every unparsed draft line read as a DEFECT
2026-09-08 13:46:38 -06:00
fe45f7f016
docs(phase-34): task 6 addendum — Drew's #ai reply posted; docs/outreach/ untracked and ignored on his decision (sent notes stay local); doc_links DEFAULT, Reference-index row, SETUP rows updated; corpus regenerated
2026-09-08 13:23:18 -06:00
c9a3d7b891
tools+docs(phase-34): task 2 part 1 — the decomp.me paste form: verbatim_target_s.py --gas (assemblable target), decompme_replica.sh steps D (the paste through decomp.me's as, 26/26) + E (the bundle), doc §5 step 2, SETUP rows, dictionary, kit corpus regenerated
2026-09-08 11:54:22 -06:00
845fd8848b
kit+docs(phase-33.5): task 15 prep — the health chain into the record corpus (14.5 addendum), the regenerated timeline, the task-15 log + the closed checkpoint
...
- Drew's question "is make tools-health set up for the decomp-architect to use as well?" — it was not: the chain's checks ship as
tools and its lessons as kernels, but the Makefile recipe (a comment per rung naming the incident that earned it) lived outside
tools/. tool_census --corpus now extracts `tools-health:` + `audit-digest:` VERBATIM into decomp-architect/corpus/record/build/
tools-health.mk (derived text; --check regenerates and compares; 92 lines); the record front page's row; the intake's Phase-5
tools column and the methodology's instruments section point at it. No installer step changed → no sixth dry-run owed; the
installable snippet (a TODO(phase-N) per rung, Step 8) is recorded for the split phase.
- docs/story-timeline.md regenerated by tools-health's report step (R75)
- CURRENT_PHASE.md: the task-15 entry — real render 32 pages rc 0; tools-health OK (5,917 lines, kit_coverage's first run in the
chain); doc_links --strict OK; audit_public OK over 6,945 paths; gate_scan PASS; the P7 walk of the six milestone items, each MET
with literal output; Ghidra stopped (R23) — and the CLOSED checkpoint (NEXT = Drew's close commit, then Phase 34 in a fresh session)
- verify: tool_census --check OK; kit_lint OK; kit_coverage OK; doc_links --strict OK; audit_public OK on the new file
2026-09-08 00:09:39 -06:00
8a698e1b79
kit+docs(phase-33.5): task 14.5 part 2 — the phase-log mining pass banked: DK-69–DK-80 from 143 worklog lessons; the S92 accelerators + decision-log entries
...
- 21 read-only Opus slices over the 26 worklogs (30,540 lines read; the three giants by line range), briefed by
.run/P33.5/log-mining/BRIEF.md with the "already banked?" grep protocol: 777 candidates, 634 already banked, 143 NEW;
every cited log line verified to exist by harvest.py and read; 142 banked + 1 dropped (the miner's own low-value verdict)
- decomp-kernels.md: twelve kernels DK-69–DK-80 (instrument blind spots; verdict staleness + the health suite; what earns
belief; denominators/units/labels; leverage vs tractability + campaign scoping; models and prompts; the unattended run;
agents and the tree; edits that keep proofs; the search harness + the compiler as evidence; maintaining the knowledge base;
hosts and services), each provenance line generated from the slices' cited worklog lines (bank.py); section 8 retitled;
Coverage "In all: DK-1 … DK-80"; every count mention → 80 (kit README, methodology, SETUP row, wiki page)
- docs/accelerators.md "P33.5 S92" (two accelerators: a distillation ships with a coverage check; the end-of-project worklog
pass recovers the fixed-but-never-generalised lessons — 1 in 5 here) → cited by the twelve kernels (kit_coverage: 59 entries,
42 cited + 15 dispositioned, 0 UNCOVERED; the matcher now requires an un-numbered entry's FULL group text);
docs/decision-log.md "P33.5 S92" (R31: the question, the measurement, the pivot, the 82%/18% why, the hindsight path)
- evidence tracked under .run/P33.5/log-mining/ (BRIEF, 21 slice reports, HARVEST.md, HARVEST_TABLE.md, harvest.py, bank.py;
a dated .gitignore block); expected-manifest +1 (docs/inherited-record.md); make kit-corpus regenerated the record copies
- verify: tool_census --check OK (358 copies); kit_coverage OK; kit_lint OK (0 leaks over 436 files); doc_links --strict OK;
wiki_render --selftest 32/0; audit_public OK on the new files; 80 DK ids cited, 0 dangling
2026-09-07 23:24:14 -06:00
0c29c9e16b
tools+kit(phase-33.5): task 14.5 part 1 — the record as the third dictionary + kit_coverage (every rule and every accelerator entry cited or dispositioned)
...
- decomp-architect/corpus/record/: the how-to (13), decision-log, accelerators, retrospective, story, wave-playbook, effort-map,
gen3-standards, gen3-handoff, DIGEST and every PhaseEnd (34) verbatim behind an authored front page (what each is, how to
read it, what is NOT there — the phase logs, R19 — and that the mining pass is their distillation); tool_census: RECORD_SOURCES
+ record_dest + the third corpus in plan/write/check (358 copies + 28 pointers, --check OK); kit_lint exempts corpus/record;
SETUP Step 6 gains 2c docs/inherited-record.md (+ the verify line; expected-manifest +1); ops-setup/README/tree/methodology/
wiki page/Home/Tools page/README bullet/SETUP row: "two dictionaries" → three
- tools/kit_coverage.py (+ config/kit_coverage_map.tsv): derives R1..R83 from DIGEST §3 (asserted contiguous) and the 58
accelerator entries (headings + numbered items), asserts each is cited by a provenance line of the registry seed / the
kernels or dispositioned (G / DK / FOLDED:G / ENV / PA / SEED: / KIT: / RECORD / COOKBOOK / NOT-PORTABLE; unknown ids
refused); first run: 26 uncited rules + 21 uncited entries → DK-66 (a ledger's tie-break, a checker's widening and a blanket
commit are part of the instrument — R70/R80/R52), DK-67 (the ignore file's directory-form wall — S91 (1)), DK-68 (a
summarised signal is a claim, not ground truth — R14/R66) in a new kernels section 8 (the museum is 9; "In all" 68) + 41
dispositions (15 PA, 3 ENV, folds into G6/G18/G38/G66/DK-12/19/20/22/25/26/31/35/44/45/46/57/61, 1 KIT template, 1 COOKBOOK);
now 0 UNCOVERED on both populations; wired into tools-health after tool_census --check; SETUP row + dictionary row
- verify: tool_census --check OK; kit_coverage OK (rules 57 cited + 26 dispositioned / 83; accelerators 41 + 15 / 58);
kit_lint OK; doc_links --strict OK; wiki_render --selftest 32 pages / 0 unlisted
2026-09-07 23:03:34 -06:00
827295e241
tools+docs(phase-33.5): task 13.5 — the tools audit + the two dictionaries: tools/tool_census.py (two agreeing enumerations of 327 tool files; docstring/SETUP row/consumers/class derived from the tree; the authored half in config/tool_dictionary.tsv — phase · portability · the NEED each tool answers · what · adapts · verdict — with coverage asserted both ways) → docs/tool-index.md (need-keyed, KEEP-GEN, Reference-index row, wiki + how-to pointers), the kit's tools/MANIFEST.md regenerated (header states live 293 + superseded 28 = 321 rows), and the two verbatim corpora in-tree (Drew, confirmed S91): decomp-architect/corpus/tools/<phase>/ (302 copies + 28 superseded pointers + INDEX) and corpus/cookbook/ (the cookbook, its symptom index, the codegen map, a front page stating what transfers per compiler) — sha1-equal to their sources by tool_census --check in tools-health, regenerated by make kit-corpus; kit_lint exempts the corpus dirs (verbatim evidence) but syntax-checks them; G66 (consult the tool dictionary first) + G67 (translate an inherited idiom through its pass) + two memory seeds (34 at install); SETUP Step 6 installs docs/knowledge-corpus.md and checks the manifest against its own stated total; the ops-setup dictionary rows; the intake's Phase 7 cites G66/G67 and Phase 10 + Part C name the raw-cast → declared-symbol step; templates/layout-contract.md (the five-tool probe, a draft for the split). The review under Drew's criterion: 93 no-consumer tools (one Opus agent's draft, verified: 0 defects, every successor live, 0 live consumers, 0 collisions; four one-off verdicts overturned to STILL-NEEDED) → 34 retired by git mv to tools/sunset/ (28 superseded, 6 one-offs; README review table; SETUP rows moved; Archive-index group). Run 4 (fresh throwaway, the final kit): stopped on my Step-6 check (321 vs the live 293) → both sides derived → resumed → PASS 10/10, manifest 56 == 56, 4 commits, guardrails held (the one foreign path was the timeline regenerated by the detached tools-health). tools-health OK; doc_links --strict rc 0; audit_public OK over 6,842 paths; the purge probe PASSED (Phase 34's gate open). decision-log "P33.5 S91" + accelerators "P33.5 S91" banked; log + checkpoint (NEXT = task 14, xHigh, fresh session)
2026-09-07 22:09:15 -06:00
9235800fb7
tools+docs(phase-33.5): task 11 — kit part 2: the firewall pack (templates/gitignore.decomp extracted byte-for-byte from the wiki fence — gitignore_template_check now runs in tools-health; firewall.txt with purge:/glob:/required:/pending:/fixture: rules; audit_public.template.py generalised from the repo's audit with its sources in the config, refusing zero sources; firewall-fixture/ = 16 synthetic bytes + sha1, the planted negative control), no-rom.template.yml, the docs/.run READMEs, ops-setup.decomp.md, bootstrap.template.sh (skeleton), CLAUDE.decomp-overlay.md (the four fail-safes + session-start extras), pa-overlays.md (7 fenced blocks: DIGEST, the 🛑 checkpoint block, the PhaseEnd narrative axis, effort rows, cookbook entry shape + triage table, wave-playbook skeleton, settings/mcp), the LICENSE/NOTICE/README/CONTRIBUTING skeletons, .clang-format + make-format.snippet.mk; tools/MANIFEST.md (325 tool files by ladder phase from one read-only survey, coverage 325/325, as Phase-N tasks); tools/kit_lint.py (fence-aware leak grep, the PLACEHOLDERS set-diff, in-memory compile / bash -n / JSON+YAML, the gitignore diff, TODO counts, coverage; --selftest = the R39 control) wired into tools-health; decomp-architect/README.md in doc_links DEFAULT; SETUP row; PLACEHOLDERS Used-in cells reconciled; make tools-health OK on this tree (detached run, .run/P33.5/tools_health_t11.log); story-timeline regenerated by the report step; log + checkpoint (NEXT = task 12, Max)
2026-09-07 19:22:51 -06:00
ae71efe56f
chore(phase-33.5): task 8 — tracked .run/ pruned to flip-ready: 218 paths untracked (172 tracked-but-ignored inertia files, 39 uncited t3s3 session logs + 5 duplicate close logs, the two fable_cd4 disassembly listings) — all still on disk; .gitignore re-includes narrowed under dated comments; tools/public_rewrite/untracked_after_rewrite.txt (audit-only sibling of the purge set, which stays byte-unchanged so gate_scan stays PASS); audit_public check 4 CONTENT (longest contiguous run of disassembly-shaped lines, cap 64; controls: 398/179 fail, the 40-line fixtures pass; 0 offenders among 6,443 paths after); runbook §11 residue paragraph; four tool comments annotated; SETUP row; log + checkpoint; P6 rules check
2026-09-07 18:04:15 -06:00
21c98ed5a5
tools(phase-33.5): task 7 — the checkers: doc_links.py six checks (links + pending; the archive refusal; the wiki-first allow-list derived from the Reference index + README; docs/ coverage 64/64; TRACKED/UNTRACKED citations by git with --disk and the '(not kept)' declaration; wiki-first warnings), wiki_render --selftest reachability (31 pages, 0 unlisted), timeline.py wired into report/audit-digest (was stale: 72 -> 73 rows), gitignore_template_check.py behind a loud skip in tools-health; the cookbook control 13 -> 0 dangling (5 archived-doc cites re-pointed to docs/sunset, 8 scratch cites declared not kept); SETUP rows; log + checkpoint
2026-09-07 17:55:59 -06:00
a0cf302e5b
docs(phase-33.5): task 1 — unblock, code side: seven tool docstring/comment cites re-pointed off docs/tool-designs and docs/hindsight-study (decision-log S61 entry, cookbook §260, how-to ch.03 Phase 2b); CLAUDE.md R1–R83, the H1 fail-safe dated to Phase 33 C3, the effort default = the DIGEST §1 doctrine; log + checkpoint
2026-09-07 17:16:17 -06:00
b46358f279
docs(phase-33): the Gen3 charter docs/gen3-standards.md (accuracy met + stricter than the field; sotn STYLE.md adopted + four rules of ours; the use-of-AI conduct; the measured gap; a definition of done), README 'Standards, and how the AI was used', docs/phase34-seed.md (RE-CHARTER, Drew 2026-09-07: P33 closes now at v1.32.0; the flip-gated remainder = Phase 34 = the flip + Gen2 exit at v2.0.0; Gen3 = Phase 35+); doc_links + SETUP rows; checkpoint §0a + header; log
2026-09-07 14:46:53 -06:00
fae69a9175
fix(phase-33): the wiki waits for the flip after all — free-plan GitHub offers wikis only on public repos (settings page text recorded); wiki_sync message + checklist item 5 corrected
2026-09-07 14:23:14 -06:00
c05c6081bd
docs(phase-33): README 'Tools from this project' section + wiki page (sidebar, Home row): xsig, the permuter driver (upstream pinned unmodified + our class swap; the only copy lives here), the codegen map, the decomp.me replica, the drafter write-up, the how-to; docs/outreach/tools-announcement.md stored for post-flip (developer's voice); checklist 5/5b (the wiki can be pushed NOW while private; xsig topics, fork deletion, the post); wiki_sync message corrected; SETUP rows; doc_links default
2026-09-07 14:20:05 -06:00
a30b80b600
docs(phase-33): E2 — docs/outreach/archipelago.md: the Archipelago note with the decomp's answers to the BFM AP world's data (dow byte 0x80078EBA CONFIRMED via func_8014B084 — our dayCounter row corrected u8->u16 and 'dow not yet separable' resolved; their patch sites + Time-Sanity hook resolved to shared engine functions by a 141-image sig scan; the town-ID patch retired as historic; memory-map Q#13 ANSWERED: BASLUS-00726MUSASHI at 0x80072DFC in EXE .data, the client's 0x072e02 fallback is +4 off); GitHub-issue text + Discord short form, plain style, SEND AFTER THE FLIP; SETUP row + E2 section; doc_links default; log + checkpoint (36 of 41: nothing left for Claude before the flip)
2026-09-07 13:29:48 -06:00
95c142d2ae
docs+tools(phase-33): E1 — docs/decompme-preset.md + tools/decompme_replica.sh: the decomp.me preset proven through decomp.me's OWN toolchain locally (old-gcc 0.13 + maspsx 86ccd7d8 rebuilt from decompme/compilers' Dockerfile; func_80018F20 26/26 words BYTE-IDENTICAL, control identical, cc1/maspsx deltas text-identical; -O1 and empty-TU negative controls; --upstream drift check); SETUP's stale 'decomp.me uses our maspsx pin' corrected (3 rows), §6.5 rewritten, E1 section + rows, ledger 14; presets are created by decomp.me's maintainers from an issue template (no UI create, no owner delete) — Drew's post-flip step = prove the scratch, file the request, the manual search; doc_links default; .run/P33/e1 log allowlisted; log + checkpoint (35 of 41)
2026-09-07 13:12:08 -06:00
ec6db382db
checkpoint(phase-33): S88 update after the outward pushes — the nine commits are pushed and both workflows ran green on them (incl. the new gccmap_cites + xsig audit steps); the gc ran (one 80 MB pack); probe_github's positive control now checks origin/main, not the local main (an unpushed commit failed it); the block's to-do, repo state and file inventory reconciled
2026-09-07 11:54:25 -06:00
Drew T