Research: independently audit PR #2588's swapped-alias frame bill and lender handoffs

[commdata2338]

Change and mathematical scope

Add research/t74-swapped-alias-frame-audit/: an independently implemented finite certificate for PR2588's two lender/helper substitutions, with a scoped analytic corollary. The construction and its improvement remain credited to the predecessors. No defect was found in the audited component. No new kappa or complete record admission is claimed.

For both pinned providers, explicit GF(2)^9 vector-set replay reconstructs the full local delta -2 e1 -2 e3 +2 e4 against PR2575. The prelude contributes +4 e1, B contributes -6 e1 -2 e3 +2 e4, and ordinary A, centre COPY, restore and final charges are unchanged. Rank mass remains 5,251 and the 422-helper inventory remains {4:33,5:9,7:62,8:78,9:240}. The certificate localizes every changed register histogram and all twelve target-room changes: eight new line-45 levels and four removed line-30 levels.

For every 0<p<1, the local moment difference is 2(4^p-3^p-1)<0. The README proves this throughout the interval, conditional on unchanged remaining charges; it does not infer feasible hosting or a global saving.

A separately written reverse-covector interpreter checks only PR2588's two newly moved handoffs: lender 80 at B916 and lender 88 at B931, in both providers. Each row is its own arbitrary input in a free algebra of invertible warp symbols. Four sign mutations preserve the entire frame receipt and break these identities.

Not checked: full scalar endpoints/cleanup-buffer values, actual native Clifford phases, physical owners/private stock, precision/tape, global or outer arithmetic, all-size realization/compiler/analytic assumptions or bit H1-H5. The selected record and frozen predecessors are unchanged.

Task and overlap

The claim is in discussion #2605. The full structured card is TASK.json.

• Task ID / scope key: complex-new-helper-inventory / 5868e5281a1e63d41a22; refined domain PR2588/PR2575/swapped-aliases/full-frame-invoice/both-providers/vector-set-audit. Nonce b3b5ee2f84cf3429266e004dacac0b21, track complex-mechanisms, priority revision 4.
• Base / exact heads: main 0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2; PR2588 8cd9dc88e24f062e2ca2fa7f4844a3d3319381fb; PR2575 921c229b4f030121a1d305de6259e174cd391955. PR2530 b4012098048e12500b2d33d2e84be281915a27ac supplies mechanism attribution, with no direct code import. The four word inputs are copied verbatim with original paths, Git blobs and compressed/decompressed hashes. Canonical main remained unchanged at the 16:24 UTC refresh; both PR heads were rechecked anonymously and unchanged.
• Concrete object / question: derive the full chronological local charge delta and width conservation; check arbitrary-dirty readiness at the two changed successor cuts.
• Scope status / reason: changed from construction to independent auditing of a changed source binding, as welcomed by priority revision 4. The original card's construction acceptance gate is not claimed.
• Exact claim audited: the swapped word's equal-width histogram and earlier handoff readiness. The inherited frame-charge contract is retained.
• Method / bounded domain: all events in four nine-bit words, 512-element subspaces, exact integer charges, four backward noncommuting rational row identities. No optimizer or packing search.
• Nearest overlap: PR2359 prices PR2322's earlier handoff; PR2472 checks entrance wrappers; PR2478/2508 guard centres/helper indices. Discussion CLAIM: PR2530 source-partner handoff cuts, independent dirty-lender audit #2606 audits PR2530's eight-source restoration/read order, so this package limits content checks to the two changed PR2588 cuts. Discussion Cost-aware source-partner helper assignments above PR2588 #2612 searches additional assignments. This account's PR2394 checks PR2339's native initial prefix. PR2621 reports a newer order/fill on unchanged PR2588 words; its global claim is not audited.
• Budget / required observable: no new physical cost or supplier saving; recover or refute the exact full local delta and equal width mass, then prove local moment dominance for every 0<p<1.
• Success / falsification / stop: deterministic certificate with frame/warp/cut and rank-preserving semantic controls; stop after this finite component audit within a 60-minute session.
• Predecessors / artifact / command: credited below; research/t74-swapped-alias-frame-audit, registered command below.

Overlap queries included the task card, 2588, helper-label, lenders 80, swapped alias and local invoice. Complete-paginated shared snapshots: entry 797 open PRs at 15:53 UTC and 1,252 Discussion records at 15:56 UTC; final observed 812 open PRs at 16:22 UTC and 1,276 Discussion records at 16:27 UTC. Expected offline warnings remain; metadata/package-note coverage is not a novelty guarantee or full diff audit.

Validation

CPython 3.12.3, standard library only:

python3 -I -B scripts/research_package.py --package research/t74-swapped-alias-frame-audit
python3 -I -B scripts/research_package.py --package research/t74-swapped-alias-frame-audit --check-only
make selected-record-check

Normal and optimized registered replay both passed:

PASS_T74_SWAPPED_ALIAS_FRAME_AUDIT words=4 providers=2 delta=-2r1-2r3+2r4 helpers=422 handoff_rows=4 frame_controls=18 readiness_controls=4 positive=4 no_new_kappa
PASS_REGISTERED_RESEARCH_PACKAGE research/t74-swapped-alias-frame-audit outcome=completed

Also passed: all four inputs compared byte-for-byte with their pinned Git objects; direct optimized replay from a read-only standalone export without Git, with all 12 exported files unchanged; package metadata/provenance; the chooser's local-ceiling positive control; selected-record bindings. No full cold supplier/native/global/outer replay or hosted CI was run. No independent human review is claimed.

Eighteen frame mutations cover reports, singular/wrong-image warps, a late handoff, non-centre cut read, target room, width inventory, removed helper and wrapper framing. Four coefficient-sign faults leave frame receipts identical but fail readiness. Positive controls use equivalent subspace generators and exact noncommuting/inverse covectors. Guards remain active under -O; ordinary replay writes no package files.

Credit

Jeremy Kothe (jeremykothe2), PR2588/2575 and the T74/entrance/native lineage; zxv, PR2530 source-partner recovery; Tekalign (tekuuu), PR2436 lifted-target aliases and prelude; Vlado Vojdanovski, PR2322 coherent handoffs; Cinna Davis (1337isnot1337), PR2226 anticipatory dirty borrowing. The source notice is retained verbatim. The distinct pricing/wrapper audits and centre/helper guards are credited, together with the original supplier/compiler authors. New here: the vector-set interpreter, reverse-covector checks, localized certificate and proof/controls.

Prepared for commdata2338 (@commdata2338), with substantial OpenAI Codex assistance. Apache-2.0.

AI assistance: submitted by @commdata2338 with substantial AI assistance. OpenAI Codex (GPT-6 Astra) did the derivation, the implementation, the checking and the writing; it was packaged and submitted with Claude Code. No independent human review is claimed.

Open next questions

See TASK.json and README. Screen one further pinned substitution using these charged traces and covectors, then perform complete native and global admission. Independently replaying the complete supplier/outer chain remains useful and separate. Discussion #2606 retains predecessor restoration/read-order work; Discussion #2612 covers simultaneous assignment search.

Workflow feedback (optional)

Exact-head reservations distinguished this audit from construction and packing. Refreshing the board exposed adjacent source-recovery work and allowed the optional semantic checks to be narrowed to the changed successor cuts. Future word changes would be easier to review with a compact per-provider transition ledger.

🤖 Generated with Claude Code

https://claude.ai/code/session_01X86sCSXdoDd61SDFfC1PKG