Skip to content

VR-28: integrate same-run kernel replay optimization - #481

Draft
afflom wants to merge 9 commits into
release/v0.4from
fix/vr28-single-kernel
Draft

afflom wants to merge 9 commits into
release/v0.4from
fix/vr28-single-kernel

Conversation

@afflom

@afflom afflom commented Nov 12, 2026

Copy link
Copy Markdown
Member

Closes #391 after the remaining acceptance work passes.

Source integration

Carries the existing reviewed fix/vr28-single-kernel preparation at 494fc28a into accepted release/v0.4 at 5dccf9e5, retaining both histories and the actual #394 landing. Published source commit: 40c9f6a586f4a78e366fd284a92d9a7deced0387; reviewed tree: b4cdbceab6687399f160b1dcd8b14b5766eb8bd5.

Generated project modules and certificates A/B/E omit their duplicate in-process kernel check only when their successful same-run replay binds the same module and complete main/server/private olean parts. One final guard covers local/imported records, the certificate tail, and staged artifacts. Runtime, support, probe, extraction, and audit modules keep their in-process checks. Existing prefix replay behavior is unchanged; this does not claim globally exactly-once declaration checking.

The only content conflict was the append-only verification history. The entire accepted history and exact VR-28 evidence suffix are retained. Independent source review found a stale capability census; SPEC, the spec-links gate, and README now consistently include all 484 IDs and VR-28. Accepted Host/policy diagnostics are preserved.

Evidence for this source integration

  • Pinned Rust 1.97.1 formatting passed on the final source.
  • cargo check --locked -p lexlean -p repo-conformance --all-targets passed in 95 seconds before the documentation/census-only corrections.
  • Static cross-check: 484 unique matching register/specification/scenario/named-test IDs and exact claim statements; both schemas parse.
  • Independent source review found no remaining blocker to draft publication. Whitespace checks passed.

The earlier controlled gateway verification pair remains 1,863.117 s to 1,243.365 s, 33.264% lower observed wall time in that single pair. Its source, host, ordering/cache limitations, and evidence are preserved in VERIFICATION.md. This is not a Rule30 speedup claim or final-tree qualification. No benchmark was repeated.

Required before readiness or merge

  • Coordinate one all-language lock/identity and generated-product wave on the final source, including model documents, fixtures, golden output, exports, and verification records.
  • Capture Atlas publication/binding and width evidence through the already-required writer and final hosted shard/join checks.
  • Run one complete final-head gate and reproducibility; review the generated artifacts and verify accepted landing/tree identity.

Generated artifacts remain unchanged and are not acceptance evidence for the new schema identities. No new Lean execution, production export, product writer, or full CI gate has been claimed for this reconciliation. Keep this PR draft until those obligations are discharged. The existing source branch remains the sole #391 carrier.

@afflom afflom left a comment

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Bounded source-integration review CLEAR at40c9f6a/treeb4cdbcea, not readiness approval. Actual ordered parents[5dccf9e,494fc28a] retain both histories and the accepted #394 landing. The entire accepted verification history and exact earlier VR-28 evidence suffix are preserved. Core replay/binding implementation and both schemas match the previously reviewed composition; runtime/support/probe/extraction/audit checks remain, and this is not a globally exactly-once claim. The484-ID register/test census and VR-28 statement mirrors are consistent.

One focused honesty log from43151978617/job130963305444 confirms R1: CONFORMANCE.md is stale versus model/*.toml. This matches the acknowledged generated-document debt; use the official projection within the coordinated product phase, not a hand edit or blind gate rerun. Repro43151978472 failed its just-repro step, but its cause has not been diagnosed; no second log was fetched or cause inferred.

The two changed schemas still require the documented coordinated all-language identity/product work, actual Atlas binding/width evidence and one complete final-head gate plus repro before acceptance. Existing source/measurement evidence is retained only for its original scope. No completed kernel-false, SP-05, VR-21/23, export, verification or timing pair was repeated by this review. Existing owner retains the sole #391 carrier and writer/qualification scheduling; #391 remains open. Independent receipt3c5be84a records the bounded evidence.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant