[llesieur99]

What

Repair retained C for the five largest available unfinished USA v1.01 routines
(58,424 retail bytes). No complete function matches, no matching credit is added,
and none of these drafts is enabled in the functional manifest.

• func_L22_0039D708: recover three float-to-int conversions, the canonical
  collision and animation callees, all five collision arguments, complete byte
  addresses, and real vector/matrix scratch storage. The old draft silently used
  zero for unsupported operations and split vector fields into unrelated locals.
• func_L11_00420C88: preserve separate projectile paths, the shared hero-state
  object, strict angle comparisons, and selector/mount transition order.
• func_L00_00300618: use an explicit C loop that preserves the four source-cursor
  advances. Its complete 24-word record-copy sequence now equals retail at all
  27 USA placements, without changing the compiler.
• func_L00_003AE900: use normal aligned template assignments with the current
  compiler, preserve the state reload following the collision call, and describe
  the actual voice-record layout. Both template-copy loops agree at all 54 USA
  v1.01/v2.00 placements; the rest of the function still differs.
• func_L08_003CDF98: correct the voice-active byte and record stride, capture
  copy inputs, restore the diagnostic argument and use an absolute hero-pointer
  load. Its inherited fast-math profile remains provisional.

Keep known helper return contracts even when ignored, and retain the command
packer's 64-bit argument widths. These were checked against the matched callee
definitions and recompiled; they do not produce a complete match.

Function	Previous draft	Retained draft	Retail
func_L22_0039D708	15,396	15,528	16,068
func_L11_00420C88	15,384	15,772	15,832
func_L00_00300618	11,272	11,280	11,348
func_L00_003AE900	8,412	8,432	8,440
func_L08_003CDF98	6,624	6,632	6,736

Size is diagnostic only. The updated source headers, nonmatching index and existing
compiler/blocker notes record the remaining differences and useful next experiments.
Production sources, flags, catalogues, the functional manifest and progress reports
are unchanged. No game data or generated evidence is included.

Checked

Validation:

• tools/build.py all --strict --version us: 28 programs OK, zero C fallbacks.
• tools/build.py all --strict --version us2: 28 programs OK, zero C fallbacks.
• tools/build.py all --strict --version pal: 28 programs OK, zero C fallbacks.
• python3 -m unittest discover -s tests: 339 tests passed.
• tests/check_repository.py and tools/progress.py --check for all three releases passed.
• All 7,567 qualified production inputs agree with the host checkout; LF/diff checks passed.
• Candidate checks: 0/87 complete placements. No functional replay or gameplay test.

Candidate checks still reject all 87 applicable placements. Only func_L00_003AE900
has the same identity in USA v2.00; all five identities are absent from PAL. This
does not claim that related gameplay code is absent from those releases.

tools/regalloc.py func_L00_003AE900 --json confirms that diagnostic and normal
objects agree. No functional capture replay or native gameplay test was run; the
strict production builds above do not qualify code retained in nonmatching/.

[bytematchdev]

ByteMatch progress report

SCUS_972.68_v2

Code matched	81.01% · 4.52 MB of 5.58 MB
Fuzzy match	81.01%
Functions	7,595 / 8,634
Data matched	0.00%
Units complete	973 / 1,156

No function changes.

SCES_516.07

Code matched	54.15% · 3.09 MB of 5.70 MB
Fuzzy match	54.15%
Functions	6,996 / 8,672
Data matched	0.00%
Units complete	958 / 1,196

No function changes.

SCUS_972.68

Code matched	84.74% · 4.68 MB of 5.53 MB
Fuzzy match	84.74%
Functions	7,760 / 8,579
Data matched	0.00%
Units complete	924 / 1,018

No function changes.

_{Commit abcc1a4 · Ratchet & Clank: Going Commando on ByteMatch}