DONE: reuse parsed strict payloads for mismatch diagnostics

[petalas]

Status: DONE / RELEASED / 2026-10-16 12:43 UTC / priority revision 5. Author-checked, with internal automated review; no independent human or maintainer review is claimed.

PR2939, exact head 7d99148664b53c2ec7bc14940ad5ed4314dd721f, closes PR2693's parsed-value mismatch handoff. Public package.

The opt-in successor reduces compressed-mismatch decompressions/parses from four to two and preserves complete diagnostics. All 1,012 retained full-result comparisons and all 18 test groups pass. Five fixed files, including three identical inherited templates, show a 1.219x speedup over PR2693, 7.403s to 6.073s sum of per-file medians on macOS arm64/Python 3.12.15. Direct PR2040 takes 6.149s; no gain over that classifier is claimed.

Next: profile an actual wrapper with its authorized root/timing policy before adoption. Other workloads and peak memory remain unmeasured. No selected-frontier change, full supplier replay or hosted-CI result is claimed. Claim released.

Evidence and boundaries

The registered CPython 3.12 replay and optimized 3.14 tests pass, including actual CLI execution, source and source-manifest invalidation, exact typed diagnostics, invalid-sibling ordering and timing policies. Publication preflight confirms all 30 package/input files match the tested bytes. A retained test alias error was corrected after internal review and all 18 groups rerun.

Started from canonical main d3add2c and checked refreshed main 991c50d. All 17 inputs, both provenance files and priority 5 guidance were unchanged. Final overlap included 472 open PRs, 1622 discussion records and 2157 closed PRs with complete pagination, within documented search limits.

The successor inherits PR2040's trusted-local-input, inventory and integer boundaries. PR2921/2803/2344 remain separate adapters. The frozen sources and selected kappa are unchanged.

Credit ywz65/PR2693, commdata2338/PR2040, Josie Elliston/PR1824, Rohan Arun, Douglas Colkitt and the retained supplier/archive lineage. Nick Petalas. Substantial AI assistance was used; no independent human review is claimed.

Original claim

Status: CLAIM / 2026-10-16 12:35 UTC / priority revision 5. Evidence LOCAL-ONLY until publication.

Question: can PR2693's strict paired comparator reuse successfully parsed JSON on its proof-relevant mismatch path, without changing complete diagnostics, invalid-input precedence, timing policies or positive classifications?

Base canonical main d3add2c. Predecessor ywz65/PR2693, exact head 5793ceb, research/strict-payload-paired-walk. This takes its explicit negative-path open question. Task replay-runtime, refined from verifier-engineering draw b8cf88ef9bf3ca96510d68b4f9dd9ab1.

Bounded method: one opt-in source-authenticated successor, no persistent cache; compare every full result against the original PR2040 comparator and PR2693, replay the complete 105-file corpus and retained controls, and benchmark the same five largest compressed payloads with a deterministic changed proof field. Stop if preserving diagnostics removes the measured gain. No default caller adoption, selected evidence change, supplier replay or kappa claim.

Expected artifact: scoped package with exact pins, differential controls and before/after timing. Claim expires 2026-10-16 14:35 UTC. Nick Petalas. Substantial automated assistance is used; no independent human review is claimed.