Finite complex construction: four exact U8 B-stage shear rewrites (no new kappa)

[martinxxx9]

This adds four exact U8 B-stage shear rewrites and a complete finite verifier. The previously submitted head omitted verify.py and its source pins; this update publishes them and reproduces the original result certificate byte for byte, addressing Doug Colkitt's review.

For distinct registers, four fragments beginning at B indices 176, 194, 204 and 278 are replaced exactly:

x += a*s; y += q*x; x -= a*s
    -> y += q*x; y += (q*a)*s

Result

• B-stage operations: 1,918 → 1,914.
• Scalar macros saved per orientation: 4.
• Forward real-component primitive bound: 193,306 → 193,254.
• Backward real-component primitive bound: 193,354 → 193,302.
• Exact arbitrary input columns: 728; paid rank profile preserved.
• Rewrite negative controls rejected: 9.
• No new kappa or selected-frontier change.

Complete replay and inputs

SOURCES.json declares the immutable input closure: the already committed research/community-frontier942-audit/t37-replay.tar.gz (SHA-256 48285025105210b56a32d3d4d57a21b7591760acc709d38073e4dca339eb06fe) and the exact hashes of its canonical word, baseline emitted programs, and baseline receipts. The canonical-word digest remains 387a2266ffe006d20042221d967f67a4b18d9bb8de1cd1e01d4031b539fa74a5.

The primary verifier uses only the standard library, reads pinned regular archive members without extraction, and imports no construction code. It checks each local identity and both complete B-stage maps with exact rational arithmetic, regenerates both baseline programs, compares the baseline receipts, and recomputes the candidate maps and primitive invoices. Its result must match the original RESULT.json byte for byte.

The package test exercises normal and optimized isolated Python, relocation with only the package and declared archive, and four rejection controls: missing source pins, mutated archive bytes, a wrong member pin, and changed result bytes. These checks remain active under -O.

Reproduce and validation

Python 3.10+ and the standard library:

python3 -I -B research/u8-b-cancellation-seed32/verify.py
python3 -I -B research/u8-b-cancellation-seed32/test_package.py

Both commands pass. Normal, optimized, and relocated results match the committed certificate exactly. The updated PR applies cleanly to main at ae95f60feec79897b153dba6459bb3f9fd4d7b8e, where the optimized package test also passes. git diff --check is clean.

The retained local emitter and independent frame/wrapper auditor were rerun on the verified candidate. Both orientations preserve the paid profile, 728 columns, and 192-bit allowance; prefix bounds are 55/72 bits and all 25 inherited controls are rejected. VALIDATION.json records this evidence and the exact validation-source hashes. The README includes the additional reproduction commands.

Scope and attribution

This establishes finite B-stage equality and local admission. It does not rebuild global hosting, the complete supplier invoice, outer inequalities, all-size/compiler or realization interfaces, precision/tape or exact recovery, or analytic transfer. It claims no new kappa, practical multiplication speedup, or formal proof of the complete theorem.

Assignment seed: 32, complex-supplier track. Original base: abdc6a7002abdbcd9e21bfc5e1a795784deacfdf.

Prepared for @martinxxx9 with substantial OpenAI assistance. The complete verifier and input pins were reconstructed and published with OpenAI Codex assistance in response to the review. Retain Jonathan Capalbo (#942), Chafik Boukhalfa (#890/#741), Jeremy Kothe (#869/#870/#799), Hao Yan (#858), icekylinx, DreamingOfClouds, and the source notices' full contributor and AI-assistance attribution. New material is Apache-2.0; no independent human review is claimed.

[Mister-Meeseeks]

100-PR batch review at c9d2523a3d30b8f6b16cf038bf723a191342ff20.

This remains open after scoped triage. Next review step: The reviewed Git head omits verify.py and source pins used by test_package.py/README. Commit the complete primary verifier and its declared immutable input closure; no test was bypassed.

This is a specific continuation point, not a rejection or a request to redo already recorded checks. Completed local evidence is retained separately from outstanding whole-head obligations.

Batch review, attribution and evidence. Thank you for the contribution.

[martinxxx9]

Thanks, Doug. Addressed in e542ad1: the complete verifier and immutable input pins are now committed. The original certificate is reproduced byte for byte, and normal, optimized, relocated, and trial-main checks pass. Validation evidence is recorded in VALIDATION.json