Audit selected generated identities: complete field coverage, no defect found

[adamgall]

Result and scope

This is an independently replayable no-defect field-coverage audit, not a repair, new supplier or new kappa. It adds only research/selected-generated-identity-audit/; no frozen source, archive, existing pin or selected record changes.

The audited comparator is research/gauge44-slots-t16/code/generated_identity.py. Inputs are the seven authenticated slots/local members of the reviewed PR1042 archive inherited by selected PR1094—not the historical R492 evidence/local directory. The exact archive bytes and member hashes stay pinned; the comparator is also bound to the selected-source manifest.

Exact evidence

• 939,603 retained tree nodes independently enumerated, including containers, keys, positions and scalar types. The original trees contain 939,604 nodes: only root phase seconds is excluded.
• All four raw-reference slots are checked before replacement by content identities.
• 155 receipt-node omission/replacement mutations, all rejected. These exhaust both receipt trees except the documented valid-timing replacement; this is not a claim of 939,603 separately executed mutations.
• 181 total negative controls, including stale references, Boolean/integer confusion, unfamiliar nested timing fields, invalid timing, lost payload fields with correctly rebound references, serialization drift and truncated gzip.
• Nine positive controls, including controlled gzip recompression and plain-JSON reformatting with updated raw recipe references.
• Five focused test groups include a deliberately faulty phase projection, exact independent-encoder bytes and immutable archive header drift rejection.

The expected canonical identities use an independently implemented recursive encoder, not the audited canonicalizer or json.dumps. An additional typed-path oracle checks both normalized receipt trees. The two gzip identities bind full decompressed bytes, not a field projection. Generated temporary containers may differ; immutable archive hashes remain strict.

Reproduction and observed verification

Base: 0b1cccf86e2f5132ed286a26ef0a681a495740ea. Python 3.12.12, macOS arm64, standard library only.

uv run --python 3.12 --no-project python -I -B research/selected-generated-identity-audit/verify.py
uv run --python 3.12 --no-project python -B scripts/research_package.py --package research/selected-generated-identity-audit --check-only
uv run --python 3.12 --no-project python -B scripts/research_package.py --package research/selected-generated-identity-audit

Publication recheck passed in 16.05 seconds combined:

PASS_SELECTED_GENERATED_IDENTITY_AUDIT nodes=939603 receipt_mutations=155 positives=9 negatives=181
PASS_PACKAGE_METADATA_AND_PINS (not a mathematical acceptance check)
Ran 5 tests in 3.892s
OK
PASS_REGISTERED_RESEARCH_PACKAGE research/selected-generated-identity-audit outcome=completed

TASK.json, SOURCES.json, VERIFY.json, RESULT.json and README retain scope, source/Git-blob pins, explicit replay, deterministic evidence, limitations and an actionable open next question.

Overlap and attribution

Assigned card: portable-replay; coordination: #1884.

The shared offline snapshot covers 543 open PRs (13:23Z) and 529 discussion records (13:25Z), with complete pagination on 2026-10-14. Targeted live checks confirmed these closest PRs remain open at the reviewed heads:

• Tooling: preserve full payloads in PR589 gzip rebuild comparisons #1559, c4f8af1dcf9fc2317c893b6b250007ded5ef879f: two PR589 gzip byte-comparison sites, not this projection.
• Tooling: selected final stage reruns byte-for-byte on CPython 3.10-3.14, reproducing the certificate bytes (no new kappa) #1824, ad5b1248f25a71983b20617e2f3a80be22f03990: selected final-stage output comparison across interpreters, not this independent complete-tree oracle.

The existing T37 role-binding audit concerns a different eight-input normalizer. Joint-prefix validation already supplies cross-platform identity agreement; this adds the bounded independent field-coverage and receipt-mutation evidence. The selected1042 lineage-resolution package is credited for locating the actual selected R480 artifacts. Search omits code diffs and most closed-PR bodies, and cached results may miss subsequent work; no novelty guarantee is claimed.

Credit Jeremy Kothe, Rohan Arun, Hao Yan and the retained supplier/compiler lineage; Douglas Colkitt's maintainer replay and identity convention; Maxime Fleury's gzip audit following Antoine Olivier's independent replay. Prepared for Adam Gall with substantial OpenAI Codex assistance. Apache-2.0; no independent human review claimed.

Limits and open next question

No supplier rebuild, selected cold replay, cross-OS run, full caller acceptance audit, execution authentication, universal JSON canonicalization result or full theorem verification. SHA-256 collision resistance and the pinned implementation/input domain remain assumptions. The inherited reader is not claimed to reject all ambiguous JSON merely because this audit's fixture reader does. Production callers may impose additional byte pins beyond the identity function.

A distinct follow-up can audit another overlap-checked normalizer against its explicit role/field contract. This package deliberately makes no broader claim. No merge or selected-frontier change is requested.

[Mister-Meeseeks]

Batch review at 7fa59a6a5bbb22e4ba6c8aabfa171245e1256cc7.

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

Reviewed scope: Independent encoder and typed-path audit binds939603 retained nodes of seven actual selected local artifacts.155 receipt mutations and181 total controls reject while9 legitimate normalization controls pass. Only root timing is excluded and references are checked then rebound; five tests pass, with no complete caller or supplier admission claim.

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.