Independently audit PR2917's paired-helper frame ledger

[ToikY]

Change and mathematical scope

Add research/paired-helper-frame-ledger/, a source-pinned independent finite frame/accounting audit of PR #2917. Its two inherited frame checks repeat essentially the same pivot-basis/climb implementation; this checker instead represents each frame by all its vectors in GF(2)^9 and reconstructs chronological paid events from the literal operations.

Both providers, all four parent/final words and nine extraction stages pass. The derived delta is +9 H1 +18 H3 −18 H4, with paid rank mass 5179 → 5170, helper-width sum 3323 → 3314, and private cleanup names 73 → 82. All 13 frame refits equal old frame intersected with the extraction hyperplane. Reverting helper499's separate scale refit makes its later rank-eight touch illegal.

Intermediate stages are frame projections. Cleanup is checked as exact inverse-reverse syntax and buffer renaming. This does not prove compensation/operator endpoints, zero-buffer restoration, native implementation, allocation, physical stock, precision, outer admission or the retained all-size/compiler hypotheses. No new κ, supplier admission or selected-frontier change.

Task and overlap

• Picker: independent-moment, scope key c6f2494b197df3bc3591, priority revision 5; refined to PR2917's new nine-step input and distinct set-based rank method. Task card.
• Entry main 8c8bcecb03def61bca701dd3b045932d39c48022; refreshed publication base 4e8ceb80ecbc11049a44856b91f74a7448a75e93. Guidance, catalogue and selected record are unchanged.
• PR2917 224bd5e56041c40078ff6b5fa7393733a215de5c; PR2908 51309592971e3b5be935599eed2a95404dc7781e; PR2894 18d03c447886895ef9abf888398178e0428533ca. Copied raw inputs match exact Git objects, including actual PR2908 WORD blobs matching carried PARENT blobs.
• Canonical claim. This complements mdumitrean's active allocation work; no allocation search. Earlier independent audits Independent verification: #2802's seven-stage increment over #2751 (retired5 + retired7 stages) passes 28/28 rules and 20/20 controls; covers #2818's identical construction (no κ) #2847/Audit #2751 independently and reconcile it with #2720's T74 centre floor (no κ; no defect found) #2865 concern different words; Exclude rank-seven helper408 extraction at the fixed terminal cut #2918 concerns helper408's fixed-cut rank-seven obstruction.
• Entry/fresh overlap covered 476/469 open PR bodies including drafts, local package notes/tasks/ledger, and relevant current Discussion claims. Non-atomic pagination, incomplete older Discussion coverage and unpublished work remain limitations.

Validation

CPython 3.12.14, macOS arm64; standard library only:

python3.12 -B scripts/research_package.py --package research/paired-helper-frame-ledger
python3.12 -B scripts/check_research_commit.py --package research/paired-helper-frame-ledger --ref 5c6c32db4449aa661aca3c37416df0d7593908bd

Registered normal and optimized replays pass; 10 test groups pass in both modes. Controls cover equal-rank wrong subspaces, chronology, invertible wrong warps, omitted terminal charges, reports, helper499's scale, cleanup aliases/inverses/types, transform binding, raw inputs and strict JSON. An independent pivot-rank oracle checks 128 small generating subsets, 16 subspaces and 256 containment pairs. A separate AI review also reproduced the section/mass calculation without importing this ledger or predecessor code.

RESULT.json stores deterministic receipts; verify.py --details emits freshly derived per-event ledgers. All 24 committed files match validated local bytes; diff whitespace check passes. Selected baseline bindings only pass. Full selected/candidate supplier replays, formal/human review and hosted CI are not claimed.

Credit

Mason Hale: nine extractions and 13 refits, PR2917. Jeremy Diviney: immediate paired-helper408/compensation architecture, PR2908. Jeremy Kothe: predecessor frame/preload chronology, PR2894/2833. The original source credit text is retained byte-for-byte, including Yiyang Lu, Dugongue, Marius-Paul Dumitrean, Philippe Van Velzen, Tekalign and icekylinx.

Prepared by ToikY with substantial OpenAI Codex assistance, including separate AI implementation, test and review agents. New work Apache-2.0; no independent human review claimed.

Open next questions

Two scoped handoffs: authenticate/replay the full operator and buffer-zero obligations; continue the existing allocation owner's source-bound native/stock/precision/outer admission. The independently checked profile is an input to those obligations, not their completion.

[ToikY]

Publication verification for exact head 5c6c32db4449aa661aca3c37416df0d7593908bd: all24 public PR file paths and Git blobs match the validated local package. A fresh Git archive was replayed from a read-only directory outside Git, with PATH containing no executables. Normal and optimized verify.py --details both pass and produce byte-identical full event JSON (SHA256 dca23f8a618d9e808fe4f574956e27ca2a7a570cf704cf95c589fb6983771144); all24 package hashes remain unchanged. This supplements the registered ten-group test runs without claiming hosted CI, which is still queued.

The canonical claim/finding is DONE and released. The concurrent allocation work cited at intake is now Marius-Paul Dumitrean's #2930, head 81f3b5d12977eba09e06fbff729167d59faabe90. Its full release has not been independently replayed in this session. This PR checks the unchanged local frame/profile input only; operator semantics, zero-buffer restoration and complete native/physical/precision/outer admission remain separate obligations. No selected-frontier change or merge is requested by this session.