Adversarial audit: selected #1094 reproduce path binds its certificate to the replay only through kappa (no new kappa)

[rtyu193-rgb]

Adversarial audit: the selected #1094 reproduce path binds its certificate to the replay only through κ (no new kappa; current certificate unaffected)

Card: supplier-boundary-audit (adversarial-audit track, seed 530ff08c…), scope refined to one obligation. Are the selected certificate's fields beyond κ the replay's output, and do they follow from their pinned receipts? CLAIM #1730. Base 0b1cccf8. Adds only research/frontier1094-certificate-binding-audit/ (6 files, stdlib only, about 5 s).

Findings

1. Invoice and outer recheck of the committed Conditional kappa 0.001672660197076918: rare-pattern T16 exchanges #1094 certificate: PASS. This applies Wronskode's frontier1042-invoice-audit-seed81 recheck, unchanged and loaded as its exact Git blob. It covers profile composition, the per-vertex envelope, nine finite guards, 7+7 bootstrap levels, 47+47 outer slacks, κ as the last grid point, and 15 mutation controls.
   • The certificate is bound by path and SHA256, not the moving pointer.
   • Positive control: seed81's published Conditional κ = 0.001672188119205421: improved T16 owner packing on the seven-slot k8 word #1042 result is reproduced exactly from certificates/checkpoints/frontier1042.json.
2. Seed81's own entry point is stale on main. make frontier1042-invoice-audit-verify fails with selected certificate path moved, because it requires the selected certificate to be Conditional κ = 0.001672188119205421: improved T16 owner packing on the seven-slot k8 word #1042's. Suggested one-line repair: read the frozen checkpoint. Not applied here; Consolidation: overlap receipt for the open selected-replay tooling PRs; #1142 fails on main and collides with #1108 (no new kappa) #1654 notes the same pattern in another capsule audit.
3. Conditional kappa 0.001672660197076918: rare-pattern T16 exchanges #1094's invoice arithmetic is unchanged from Conditional κ = 0.001672188119205421: improved T16 owner packing on the seven-slot k8 word #1042. The receipts differ in exactly 14 fields: three provenance hashes in the invoice, plus the idle-remainder profile and three counters in the owner receipt.
4. The profile agrees across five artifacts:
   • the certificate;
   • the archived owner receipt;
   • the package's PROFILE.json;
   • the package's data/PLAN.json;
   • INDEPENDENT1094's owner-audited profile.
5. Binding gap, demonstrated with a well-formed mutant certificate. The mutant takes complex_lower + 1e-24, the next grid point, and recomputes the bootstrap, all 47 outer slacks and both complex moment records with seed81's formulas. κ is unchanged.

Layer	Mutant
Invoice/outer arithmetic (endpoint as data)	accepted
verify_selected_result.check with certificate_sha256 re-pinned	accepted (a κ edit is still rejected)
scripts/verify_frontier1094.py	no reference to the #1094 certificate, the selected record or certificate_sha256
Candidate stage complex-t16-rare-pattern-exchanges/verify.py	binds only status, κ and the replay's profile (to PROFILE.json); never names complex_lower, bootstrap, outer, moments
Exact moment (selected-suppliers-de-diagnostic enclosure)	rejected: residual lower bound 1.62·10⁻²⁴ > 0 (certified endpoint: upper bound −1.30·10⁻²⁴)

So the record's hash pin says which bytes are selected, and the cold replay says which κ is reproduced. Nothing on make frontier1094-verify says those bytes are the replay's output.

This is not a defect in the current certificate. The archived replay output candidate/strict/ASSEMBLY.json and the published RESULT.json assembly_sha256 both equal the certificate (1122cb3a…).

Proposed handoff (not applied: selected reproduce script, maintainer decision)

require(sha(work/'candidate/strict/ASSEMBLY.json') ==
        read(ROOT/'certificates/selected-result.json')['certificate_sha256'],
        'Replay output is the selected certificate')

Alternatively, use a parsed-JSON comparison such as #474's check_selected_replay.py helper. The package shows the requirement holds on the published replay output and rejects the mutant.

Statically, verify_frontier1042/942/890/799.py also contain no reference to their certificates. They were not audited further.

Not checked

• Cold replay: no fresh cold make frontier1094-verify was run, with or without the line; it needs the long Conditional κ = 0.001672188119205421: improved T16 owner packing on the seven-slot k8 word #1042 baseline.
• Inherited obligations: semantic receipt content (words, owners, hosting) is taken as data, and the native, all-size, precision, analytic and H1–H5 obligations are inherited unchanged.
• Environment: run on Windows with CPython 3.14. The registered 3.12 replay was not run locally; --check-only and the research tests pass.

Credit and related work

• Bind the selected gen4 certificate to fresh replay outputs #474 (artuskg, open) first identified this class of gap for the historical gen4 record. No priority is claimed for the pattern.
• Research audit: the selected #1094 reuse replay discards its ancestry recheck; binding certified from committed archives (no new kappa) #1277 (merged) covers this replay's ancestry binding; Consolidation: overlap receipt for the open selected-replay tooling PRs; #1142 fails on main and collides with #1108 (no new kappa) #1654 (open) is its tooling-family overlap receipt. Neither adds a certificate comparison.
• Reused unchanged: Wronskode (seed81 recheck) and Navinor (DE moment enclosure).
• Selected scripts and composition: Douglas Colkitt.
• Construction: Rohan Arun (Conditional kappa 0.001672660197076918: rare-pattern T16 exchanges #1094), with Jeremy Kothe's k8/T16 construction and the credited supplier lineage.

Substantial AI assistance: Claude Code (Claude Opus 5.5), one donated session, no subagents, no independent human review. Apache-2.0.

🤖 Generated with Claude Code

[JosieElliston]

A complementary data point for the final stage, from #1824 (session 26, no new kappa). I reran research/complex-t16-rare-pattern-exchanges/verify.py from the hash-bound archived #1042 slots/local on CPython 3.10.21–3.14.7, with several hash seeds. Every rerun's strict/ASSEMBLY.json was byte-identical to community-frontier1094-audit/ASSEMBLY.json (1122cb3a…), the certificate that the selected record pins. So rerunning that stage binds the certificate to the replay by bytes, not only through kappa. #1824's verify.py checks this explicitly. I made no change to verify_frontier1094.py, so the gap you describe in the reproduce path itself remains.

[Mister-Meeseeks]

Batch review at 4112c4a3c2c2f171ef1cc71a5dc422c40cd238fb.

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

Reviewed scope: Pinned invoice/outer and exact-moment mutation audit separates correct published certificate bytes from a missing certificate-to-cold-replay comparison. The historical script and candidate path were inspected; current archived output agrees, and the proposed binding rejects the mutated endpoint. This package does not install the repair or rerun the cold pipeline.

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.