Research audit: verify all selected cold-wrapper amplitudes (no new kappa)

[BigCactusLabs]

Change and mathematical scope

Research audit / independent finite verification; no new kappa. All 14 distinct four-dimensional cold residual wrappers in the selected PR626 construction pass an independent exact amplitude check on every input basis column. They occur at 44 helper instances across the CK and JP8 providers. The checker establishes 7,168 column identities, determining all 3,670,016 matrix entries, without importing the producer, Clifford tableaus, lower_child, or the source state-vector executor.

The concrete adversarial control right-composes each wrapper with Z_6. Width, all support patterns and the zero-input column stay unchanged, while exactly 256 columns change sign. At the seed-selected frame, input 64/output 11 changes from 1/4 to -1/4. This exposes the limit of those reduced checks; the existing production predicate already rejects the mutation through its tableau check. No production defect is alleged. All 70 phase, translation and order controls are rejected.

Only a new research package and its scoped workflow are added. The selected pointer, accepted suppliers and upstream/ are unchanged. The package includes the kernel derivation, exact deterministic receipt, semantic artifact/source pins, local claim, attribution and validation record. The finite operator identities apply to arbitrary complex inputs by linearity; the full scalar word, wrapper placement, hosting, all-size/compiler, precision/tape and analytic interfaces remain inherited obligations.

Base: 66849e1475c930fd7732b1a0067ffee7e1e95f60. Selected PR626 source: 660de893d27bb7abd1d5b01b1640207528a5c851; native PR622 source: f69477f4f6d3c7eb1f740b166881bffed4ffa9f6. The archived inputs match PR626's exact SPEC semantic hashes.

Validation

python3 -B research/cold-wrapper-amplitude-audit/verify.py
python3 -I -O -B research/cold-wrapper-amplitude-audit/test_verify.py

Expected: PASS_COLD_WRAPPER_AMPLITUDE_AUDIT: 14 frames; 44 helpers; 7168 columns; 3670016 entries; 70 rejected mutations, followed by six passing failure-detection tests.

• Same complete receipt on Python 3.11.15, 3.13.16, 3.14.8 and 3.15.0; isolated optimized execution passes.
• Six harness tests cover relocation, changed gzip containers, corrupt source/artifact/receipt rejection, and output preservation.
• Original helper-interface checker passes on both providers; its full predicate separately rejects all 14 new Z_6 mutants.
• make selected-record-check, actionlint and staged whitespace checks pass.
• Scoped Linux CI passed on Python 3.11 and 3.14 at submitted head 27db45ac6befa99d6c06be920a93845023e81937 (attempt 1): exact receipt, all six tests and clean-tree checks.
• Full regenerating make verify / frontier626-verify were not run locally for this isolated audit; no full-suite success is claimed. Upstream PR workflows currently report action_required for the fork contribution; fork CI success is separate from upstream approval.

Assignment seed 60, adversarial mathematical audit. No pinned sprint issue existed, so the claim is local. The overlap refresh covered 518 open PRs through #698. PR662 optimizes the original Clifford engine on the same archived inputs; this contribution instead uses direct all-column kernels. PR581 is a distinct rank-one Lean identity; PR675 covers owner/controller boundaries. None is an implementation dependency.

Prepared for BigCactusLabs with substantial OpenAI Codex assistance. Credits to chafreaky, jeremykothe2, DreamingOfClouds, icekylinx and the retained supplier/compiler lineage are preserved in NOTICE.md. Apache-2.0; no independent human review or full theorem formalization is claimed.

[Mister-Meeseeks]

100-PR batch review at 27db45ac6befa99d6c06be920a93845023e81937.

Merged to main, preserving the original contributor commits and package attribution.

Reviewed scope: Independent complete amplitude replay of 14 cold wrappers and a phase-error control invisible on the zero column; no new supplier admission.

The pinned scoped command passed; the batch ledger contains its command, input hashes, environment and log. This incorporates the research/tooling contribution and does not change the selected κ or discharge the inherited multiplication-theorem assumptions.

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