Consolidation: overlap receipt for the fourth100 witness-MANIFEST pin family (#636/#638/#639/#640/#642/#644/#646) — one drifted larger_gain pin, partial retention, credit intact (no new kappa)

[navinor]

Card: overlap-receipt (consolidation track, priority revision 3, chooser nonce 6bc45fb7886eb99e0494a94781da0e58, scope key 757723e3fd38f20fa774), scope refined to one small named PR set. Base main 1f7a934975e5eacaf83511c9ac870263bf095e05. No new kappa, no construction, no disposition changes; all seven PRs stay open and the selected frontier is unchanged.

Object: the seven entries of the fourth100 reconciliation whose contributed-package replay fails at verify.py line 39 with AssertionError: larger_gain manifest pin — #636 , #638 , #639 , #640 , #642 , #644 (vectorsss / Chi Zhao) and #646 (jeremydiviney). This is a source-bound disposition receipt: a finite certificate of git-blob identities plus one diagnosed source-pin defect, auditing the existing fourth100 receipt (research/community-reconciliation-fourth100-audit/, docs/research/community-reconciliation-fourth100-review.md, docs/research/pr-ledger.json).

Findings

F1 — the seven failures share one cause: a drifted larger_gain pin. All seven SPECs register larger_gain with manifest_sha256 = 54d67f45ad520f4a134e069336bd0ff639aaeada28d13539298e4f2655c9692d, which is exactly research/larger-gain/MANIFEST.json at the pinned PR 569 revision 897213a855b7fb8a1b8b0e552a2f5c42e6c1b326. The same path as staged into main (commit df417acd, "Stage frozen PR family through 588") hashes 73c186c2807c703e43293c7cf785557a046958c245fa812e433f7fe208cdef6d, and this drifted copy is what all seven recorded heads carry in their base trees (checked at each head via the contents API). The two manifests differ in exactly two files entries, NOTICE and README.md; both are 40,559 bytes. The staged package is self-consistent at its own revision (main's NOTICE/README.md hash to the staged manifest's values). The other five pinned sources (multicold, structural_gain, joint_prefix, frame_coalescing, direct_delivery) still match their pins in the live repository packages.

F2 — the six vectorsss packages vendor the pinned revision byte-exactly. For each, every source pin resolves to its own vendor/<name>/MANIFEST.json and matches its registered digest in raw bytes (zero CRLF anywhere; the repository's text normalization is not involved). All pinned files match too. Run through their own verify.sh source overrides, the six pass the pin block the reconciliation recorded as failing; they evidently ran with sources resolved to live repository packages (as PR 646's defaults do), where larger_gain is the first failing pin. PR 646 has no vendored copy: its default resolution of larger_gain is the live research/larger-gain, which reproduces its failure on any checkout of main.

F3 — retention is partial and dominated by re-vendored predecessor content (exact blob identity against main at 1f7a9349; no path matches at the same path and no path exists in main with different content):

PR	changed paths	byte-retained in main (other path)	not byte-retained
PR 636	348	285	63
PR 638	347	281	66
PR 639	340	281	59
PR 640	339	282	57
PR 642	354	281	73
PR 644	355	280	75
PR 646	186	77	109

The unretained remainder is each package's own witness/tools/data/result files (path lists in evidence/pr{N}-audit.json).

F4 — the six vectorsss members share their vendored core (250-268 shared blobs pairwise; PR 646 shares 31-33 with each): one content family plus six small per-PR deltas.

F5 — predecessor credit is present. CONTRIBUTORS.md credits vectorsss for PR 636/PR 638/PR 639/PR 640/PR 642/PR 644 (line 835) and jeremydiviney for PR 646 (line 833). No missing attribution found for this family.

Delta and evidence

New package research/pr636-family-witness-manifest-receipt/ only (plus the local task card notes/task-card-agent-22.md): README with Open next questions, TASK.json/SOURCES.json/VERIFY.json metadata, RECEIPT.json, per-PR evidence (evidence/pr{N}-audit.json: pin tables and every changed path with its blob identity and class), two byte-exact MANIFEST.json evidence copies, and tools (tools/audit_witness_pins.py fetch-side, tools/build_receipt.py). No existing file is modified; upstream/, frozen sources and pins are untouched.

Checks

• python3 -B research/pr636-family-witness-manifest-receipt/verify.py passes: drift identity (entries NOTICE, README.md), all per-PR pin verdicts and retention classes re-derived against git ls-tree -r 1f7a9349 (2,269 changed paths), cross-PR shared-blob counts, one positive fixture and four failure controls (tampered digest, substituted manifest diff, omitted row, substituted class), each rejected.
• python3.12 -B scripts/research_package.py --package research/pr636-family-witness-manifest-receipt → PASS_REGISTERED_RESEARCH_PACKAGE ... outcome=completed; --check-only → PASS_PACKAGE_METADATA_AND_PINS.
• python3 -B scripts/research_baseline.py → PASS_RESEARCH_BASELINE_BINDINGS_ONLY.
• python3 -B -m unittest discover -s tests -p 'test_research_*.py': 15 passed, 1 skipped, 5 errors in test_research_onboarding.PackageTests.setUp (Windows CRLF fixture: write_text CRLF vs normalized blob in a temp repo). The same module errors identically on a pristine git archive extract of the base commit, so these are pre-existing environment failures, not caused by this change.

Limitations (not checked)

Head-side blob identities rest on the GitHub pulls/contents APIs at the recorded head SHAs (PR trees are not fetched locally; verify.py re-derives the main-tree side). The multicold full_tree manifests were not recomputed file by file. The seven packages' full contributed replays were not executed; this receipt resolves the recorded pin failure and the retention/credit question only. That the fourth100 harness resolved sources to live packages is inferred from the uniform larger_gain manifest pin assertion plus the pin data; the harness invocation was not recorded. Retention matching is byte identity.

Open next questions (in the package README/TASK)

1. What source resolution did the fourth100 harness pass to these verify.py files? Re-running PR 636's verify.sh --check-pins at recorded head 679621828a85bac68af869fb0e55b54508e6b14b would settle whether the six are re-run conditions rather than content defects.
2. Should main reconcile the larger_gain pin drift (vendor the pinned PR 569 revision, or re-pin to the df417acd-staged manifest)? Every live-package resolution fails until then. Maintainer decision; this PR changes no pin.
3. Classify the 57-109 unretained paths per PR into method code worth retaining versus generated witness output (bounded, offline).

Task card and scope

The full card is in notes/task-card-agent-22.md; scope status refined with reason in TASK.json. Overlap: the concurrent selected-replay-tooling receipt claim (#1166 (comment)) and the PR 99-to-PR 103 retention claim (#1166 (comment)) cover different families; #1572 audited PR 148's path gaps and is the methodological predecessor. A refreshed find_overlap.py pass on the concrete object terms ("larger_gain manifest pin", "witness MANIFEST", "fourth100", "pr636-family") found no receipt in progress for this family. Coordination claim: #1166 (comment) .

Credit: vectorsss / Chi Zhao and jeremydiviney / Jeremy Diviney for the seven packages; the fourth100 reconciliation authors for the audited receipt and recorded evidence; the PR 569 and df417acd staging authors for the two manifest revisions compared. Prepared by Aaron Wang / navinor, with help from opencode. Substantial AI assistance disclosed; no independent human review claimed. Apache-2.0.

[Mister-Meeseeks]

Batch review at 2fa58df219b90bbca464466d2c69de08e8961862.

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

Reviewed scope: Historical source-retention and manifest-drift receipt for seven PR636-family heads: exact main-side byte comparisons, six vendored alternatives and disclosed head-side API evidence. Does not change those PR dispositions or silently repin attributed source archives.

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.