Independently audit staged544 terminal-to-owner cost bridge

[Clear1nsight]

Result

Independently corroborate PR3945's terminal-to-owner cost bridge at 91f81d131252db704f945891da1ea54e29d46b8b. Both providers freshly reproduce the local invoice; all 1,321,920 literal roles and 725,760 physical bins are checked. Exactly 84,240 bins absorb one promoted terminal dimension. The complete paid delta is 51840 H1 - 42120 H2 - 19440 H3 + 22680 H4, with zero rank-mass change and the claimed complete profile {1:8041680,2:2306880,3:729648,4:353160,5:69120,8:25920,48:137952}. No discrepancy was found in this scoped bridge. No new construction or kappa is claimed.

The checker imports no predecessor code. It uses explicit binary point sets for the full local invoice, exact interleaved free-program identity, reverse source-window queries and coordinate bitmaps for physical bins. It connects widths derived from actual terminal/entrance subspaces to the actual dispatch and idle remainder bill; it does not trust word report histograms. Exact program identity establishes relative abstract scalar/permutation equality on arbitrary dirty inputs and private buffers. Native Gaussian/Clifford phases remain excluded from that statement.

Reproduction and validation

Only research/staged544-terminal-owner-bridge-audit/ changes: 20 files, 359,268 bytes. Default registered replay retains four exact words (294,388 bytes) and freshly checks both local providers plus 11 semantic rejecting controls in normal and optimized Python. It deliberately prints PASS_LOCAL_BRIDGE_ONLY: a default CI pass is not a fresh literal-owner audit.

The full owner replay uses 12 explicit immutable public inputs fetched into a fresh directory by fetch_sources.py; every byte is SHA256-pinned and checked again by the verifier. Both final full runs passed: 62.033 seconds normal and 50.446 seconds optimized on Python 3.12.3, with exact retained-result equality and all source bytes unchanged. All 17 mathematical controls reject, including six through the actual owner production audit: changed identity, family collision, closed window, shifted offset, missing dispatch and complete-bill mismatch. Four additional production CLI controls reject a corrupted source and stale retained receipt in normal/optimized Python. Official registered local replay passed in 5.100 seconds. Commands, hashes, scope and complete cost-status ledger are retained.

This is a deliberate independent audit of Dugongue's changed-terminal construction. PR3941/3942/3957 fixed-exit retiming results do not admit that change. Canonical main 20ecf2891d80b5acf18d980b2164fb44ea7faa90 and priority revision 6 were refreshed unchanged; bounded Discussions and complete 662-open-PR-body pagination were checked. Active audit claim. No exhaustive novelty claim.

Fresh native phase recipes, complete ordinary/outer release, a new root calculation, all-size compiler, physical hosting implementation, precision/stock/tape/recovery, analytic and BIT assumptions remain unverified here. Unknown costs are not zero; the selected frontier is unchanged. The next bounded audit is the changed native phase/interface and its source-bound release admission.

Credit

Chris R / Clear1nsight with substantial OpenAI Codex assistance. The construction belongs to Dugongue/PR3945. Preserve dataisfire PR3734, zxv PR3682, Ross Richardson PR3605, Erwan Secousses/ChaloupeMusic PR3815 and retained PR3638, hexapode PR3802/3875, Sagar Thakrar PR3687, roger-sato PR3693, Arham Tahir, Jeremy Kothe, Lyy-iiis, philippeVV and the retained detailed predecessor notices. Apache-2.0. No independent human or proof-kernel review is claimed.

Frozen publication checks

GitHub-authored scientific commit 6ff67539f1131a287eb9a9a98bf1e94a02eecba7, one parent at canonical main. All 20 committed files match local validated bytes, and the declared Git provenance passes the source-history gate. The official committed package export passes integrity checking; its Git-free local replay passes in normal and optimized Python with unchanged exported bytes. Hosted CI status is separate and is not preclaimed here.

[Clear1nsight]

FINDING / scoped DONE — independent native quadratic-phase audit; claim released.

2026-10-19 08:04 UTC, priority revision 6. Published PR3987, exact scientific commit fc536202fdaa924158789a5ef59ac079bbe21922.

All promoted cold terminal residuals and five genuinely changed warp interfaces per pinned PR3945 provider pass independent exact signed verification: 40 unique interfaces, 144 source occurrences, 20,480 complete basis columns. Fourteen phase/rank/basis/coverage controls reject. Ordinary and optimized registered replay, exact committed-byte/source-history checks, and read-only isolated Git-free verification pass. Hosted CI pending when this finding was posted.

The contrasting finite-symplectic/quadratic-refinement method also exhibits a useful actual lift defect: G=[1,2,8,32,44,128,59,66,489] satisfies K_(G inverse) K_G = Z_256, flipping 256 basis phases despite an identity binary map. The author's correct adjoint passes; this is no author bug and no new κ. The correction is a nonconstant Pauli character.

Frozen word/native inventory/release crosslinks pass from included raw bytes. The complete native recipe payload, all changed scalar frame transitions, full emitted program, full release execution and uniform supplier obligations remain outside this certificate. Next useful task: materialize the full recipe payload, match retained hashes, and independently check the changed scalar transition classes. Unknown costs stay explicit in the ledger. Credit and exact predecessor lineage are retained in the package; substantial OpenAI Codex assistance. No continuing claim held.