Rebind central-factor-packed pins so community-batch952 can admit sou…

[unpaidmoon]

Duplicate of #4110. kumagi's PR #4110 repins the same five central-factor-packed sources with the same receipt 2aabc7e3… and was opened first. This PR is an independent reproduction only; reviewers should prefer #4110. See #1166 coordination note.

Change and mathematical scope

Tooling pin refresh. No new κ, no supplier, no selected-frontier change. Selected κ stays 0.001672660197076918.

research/central-factor-packed/verify.py refused current main before any coefficient check. admit_sources() still pinned five files from before the #2194 production wiring and the related #2001/#2054 tooling merges. .github/workflows/community-batch952.yml runs that verifier in the same finite job as every research/r439-* pull request, so the stale pins turned the job red even when the PR's own replay passed.

Five live pins in SOURCES.json now match canonical main bb4bccbf42844d2801487450727c2f63e10d1d0f. The other 38 pins and the package base 47481c89fd30a43f8f882eb4b391e89c29273fd0 are unchanged. The complete packed replay then matches certificates/dyadic-central-factor.json byte for byte (receipt_sha256 2aabc7e3ed38a27902f88244b80bed579e8ec2158c6ad3253f1c654c79b299e0, coefficients_checked 6760000). That certificate differs from the package's original receipt ffae0f0a22863718e3134a924592dcf4b05e919223323ba536c587c592bfe6a1 only in three embedded source_sha256 entries:

• scripts/exclusion_circuit.py
• scripts/experiments/complex_centers.py
• scripts/experiments/disjoint_tensor.py

scripts/shared_point_circuit.py also had to be re-pinned; it is not a certificate source. VALIDATION.json and BENCHMARK.json stay the October 13 record against ffae0f0a.

Task and overlap

• Base: bb4bccbf42844d2801487450727c2f63e10d1d0f (canonical CrocSwap/integer-mult-bounds main; priority revision 6).
• Task: verifier-engineering, responding to the open CI finding on Handoffs and requests for independent checks #1166 rather than a new catalogue draw. Scope refined to the five drifted central-factor-packed pins. The random task picker was not used.
• Finding: Handoffs and requests for independent checks #1166 (comment)
• Completion posted: Handoffs and requests for independent checks #1166 (reply in thread)
• Not redone: open perf(research): compile the P112 ten-deletion census with popcnt (2.69x, byte-identical receipts) #4090 (P112 census -mpopcnt, the replay-runtime lever already taken) and research/central-factor-production/ (the Tooling: 44× faster production verify_factor via packed-row equality with identical receipts #2194 wiring this refresh admits, not a second algorithm).
• No open PR title matches central-factor-packed, community-batch952, or dyadic-central-factor.

Ordered eligible list used for this session (session 3), after dropping claimed, delivered, and open-PR work, and after dropping the flag-9 head-step repair in #4107:

1. Typed actual-width physical H / rounded return — Handoffs and requests for independent checks #1166 comment 19016699 (uniform-transfer-audit). Left for another session.
2. Independent review of the tensor-pairing embedding — Handoffs and requests for independent checks #1166 comment 19030180 / PR Prove sharp common-E8 embedding bound for tensor pairing #3928. Left for another session.
3. This item: stale community-batch952 central-factor pins — Handoffs and requests for independent checks #1166 comment 19046528.

Priority-table next decisions were ineligible before that list: #3259 review delivered; flag-9 head-step is #4107; joint-query producer work is in open PRs including #3181; selected-stock binding is open #3898; the V45 ledger audit is open #3911; the scatter mutant is open #3801; the 01021323 / L22 prune work is in open PRs including #3821 and #3708.

Validation

CPython 3.12.3, repository root:

python3 -B research/central-factor-packed/verify.py
python3 -B research/central-factor-packed/test_packed.py
python3 -O -B research/central-factor-packed/test_packed.py

• verify.py: PASS_CENTRAL_FACTOR_PACKED, scope complete_historical_audit, coefficients_checked 6760000, receipt_sha256 2aabc7e3…, 1.901 s.
• test_packed.py: 11 tests OK in 3.656 s, including 1,344 scatter replacements and 231 gather sign flips and deletions.
• python3 -O -B test_packed.py: 11 tests OK in 3.630 s.

Not run: hosted community-batch952, full make verify, the selected cold replay, Lean, Docker. The certificate comparison is a byte match of the regenerated historical audit, not a new proof of the central factor.

Credit

Nick Petalas for the CI finding. Ross Richardson for the packed adapter. Wentao Li for the #2194 / #2001 / #2054 production tooling whose bytes these pins now admit. Douglas Colkitt for the dyadic central factor. Prepared by unpaidmoon with substantial Cursor Grok assistance. Apache-2.0. No independent human or formal-kernel review is claimed.

Open next questions

1. Should community-batch952.yml keep running the historical packed adapter on every research/r439-* change, or should that job call research/central-factor-production/verify.py and leave the historical pins frozen at 47481c89? This PR chooses the pin refresh the finding named, so the existing job can pass.
2. scripts/shared_point_circuit.py is admitted by SOURCES.json but is not in the certificate's source_sha256 map. A later edit of that file will again fail the r439 job without changing the central-factor receipt.

Workflow feedback (optional)

The cloud agent's GitHub integration could not write upstream Discussions or open the upstream PR (Resource not accessible by integration on the one Discussion attempt; no retry). The account owner posted the completion on #1166 and opened this PR from the fork. Overlap was checked with read-only gh searches and the open-PR title list before the pins were edited.

[kumagi]

Overlap coordination + shared evidence — PR #4110 (independent, from the same petalas FINDING) repins the same five research/central-factor-packed/ source pins and its local replay reaches the same PASS_CENTRAL_FACTOR_PACKED with byte-equal receipt 2aabc7e3…. Reviewers can take either diff; mine is data-only (5 hash values), yours adds README/VALIDATION.md narration — both are consistent with the regenerated certificate being faithful (I also confirmed central_factor, certified_complex_saving, retained_kappa and upstream_commit adc7f124… are byte-unchanged between the recorded and current certificates; only the three embedded script hashes moved).

Additional blocker your PR will hit after the pins admit (.run evidence from my branch, run 39354719970): the finite job fails earlier in Selected record and review ledger integrity — scripts/verify_pr_reconciliation.py:108 KeyError: 'closure' on clean main bb4bccbf, for the first of 24 rows without the closure key in docs/research/pr-intake-decisions.json (#1365, #1494, #2361, #2832, #2963, #2985, #3012, #3029, #3031, #3039, …). The authoritative pr-ledger.json carries "closure": null for exactly those rows (canonical generator form, scripts/reconcile_pr_queue.py:187), and 20ecf2891's "Close queue pass" left the decisions projection key-less. That step runs before your package steps, so the job stays red until the closure-projection divergence is repaired as well — it is a maintainer ledger-contract decision, so I deliberately left shared scripts and the ledger untouched.