Skip to content

Enable owned inspection and bounded certificate validation - #482

Draft
afflom wants to merge 6 commits into
codex/rule30-native-v05from
codex/rule30-kernel-inspection-v05
Draft

afflom wants to merge 6 commits into
codex/rule30-native-v05from
codex/rule30-kernel-inspection-v05

Conversation

@afflom

@afflom afflom commented Nov 12, 2026 •

Copy link
Copy Markdown
Member

Adds owned, bounded kernel-type inspection for the native Rule30 solver and partitions large preservation certificates into units that can be elaborated and kernel-replayed within unchanged child limits. Head: 6a7668285b34ea99f8b32f5cf743b44b07a1c505. Stacked on #439, following release/v0.5.

Inspection uses the embedded LexLean-authored adapter and preserves structural kernel data, request coverage, order, and tool identities. Requested names are constructed structurally: Lean's checked name quotation previously rejected an absent declaration before inspection could run. An absent declaration now retains its requested row with empty universe parameters, a null type, and an explicit native unrepresentable reason. Present declarations in the same batch remain inspectable. Missing names consume the shared traversal budget; unresolved terms, exhausted budgets, malformed output, and process/tool failures still refuse the complete report. Inspection never supplies a proof-acceptance receipt.

Preservation certificates are partitioned only at existing core-command boundaries. Every part is elaborated and replayed; the original wrapper imports the chain for the unchanged exact-axiom audit. Parts live outside original certificate prefixes because leanchecker also selects descendant modules. Failure attribution distinguishes root R14 from R1.

The clean CI build exposed an earlier local compiler identity mismatch: generated language/lcnf-1.2/tooling/lexlean.lock and .lexlean/.lock files had entered the local embedded semantic tree. The corrected compiler was built from a clean isolated source tree. All 906 embedded semantic input paths and bytes were compared with the final committed tree, and 269 tracked language-1.2 locks were refreshed with compiler-identity-only changes, preserving deliberately invalid negative fixtures. The build script's semantic inclusion rules were not weakened. The final language-1.2 identity is 79cf6fea059902c0b9f46736a3543edf29a059fdd3cc47690012b223126fc0ae; language 1.0 and 1.1 identities remain unchanged.

Language 1.2 also admits exactly LexLeanRuntime.utf8Encode and String.toUTF8_eq_toByteArray as fixed simplification bridges. Naming either selects simp only followed by reflexivity on remaining goals. Root qualification is confined to those simplify references; ordinary member rendering and historical simplify output remain unchanged. No arbitrary foreign helpers, tactic strings, proof-sequence syntax, or runtime semantic changes are introduced. This supports the actual Rule30 catalog UTF-8 identities under Lean 4.32.1.

Completed validation:

  • Current head: Lean-backed SM39 verifies empty, ASCII, and multibyte UTF-8 equalities with exact axiom policies; rejects a false byte equality through verification; rejects foreign helpers, malformed lists, language-1.1 uses, and reserved-name capture. It also checks unchanged historical simplify output.
  • Current head: RP10, model/specification-link checks, workspace all-target/all-feature Clippy with warnings denied, formatting, exact identity-only lock audit, representative compiler lock --check, and focused source review. All 906 semantic input paths and bytes match the final committed tree.
  • Base 40200f4dd5: all 225 compiler library unit tests and Lean-backed VR28, including present/missing/present inspection and absent-name budget refusal. Those tests were not redundantly rerun for the isolated UTF-8 proof-bridge correction.
  • The missing-origin regression first reproduced checked-name quotation failure, then passed after structural-name construction. The UTF-8 regression first reproduced linker refusal with the prior compiler, then passed with the fixed bridge recipe.

Earlier certificate diagnostics checked all 745 controller declarations; all 15 command groups elaborated and replayed within the unchanged 600-second child limit, and the isolated final wrapper passed. Rule30 subsequently completed its full project gate with the preceding partition compiler. Those results do not establish full upstream acceptance for this new head.

Full upstream just vv acceptance remains incomplete. The previous full run stopped at a redundant clone, then at an example-lock mismatch; those defects were corrected, but all remaining full-suite stages have not been completed. This PR remains draft pending complete upstream acceptance.

@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.

[P2] Resolve the VR-28 collision before further identity/product qualification. The new inspection claim in spec/src/IdsVR.lex.tex is a different capability from VR-28 already explicitly reserved by #391 comment6605358807 and now carried by #481 at40c9f6a. Both would occupy the authoritative/projected register, scenario, named test and dispatch after forward integration. The current #439/v0.5 registry stopping at VR-23 explains why isolated validation does not detect the queued collision.

Coordinate a free identifier for this new inspection claim with the existing #391/#481 and #439 owners, preserving VR-28's prior reservation and #362's VR-24–27 range. VR-29 was absent in the bounded register/reservation search, but this review does not assign it. Make the rename in authoritative source and use official projections for mirrors, then update the scenario/test/dispatch and PR description together; no competing patch or lowered count is requested.

The inspected subset preserves structural names/universes/binder visibility and treats inspection as data, not proof acceptance. That is a bounded source observation, not complete inspection or final qualification clearance. The draft stacked run skips required groups, separate honesty passed and repro failed; no runtime, stopped #431/TC-03 experiment, log download or new gate was launched by this review. Existing owner retains the implementation and #439 dependency/retargeting plan.

@afflom afflom mentioned this pull request Nov 12, 2026

@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.

Exact-head CI43163525764 and repro43163526076 both terminate on examples/collections. The saved property-job131000169140 log identifies formatter_is_idempotent_on_every_example failing with LLC0102 and the explicit note that compiler semantics changed since locking; lock.rs emits that note for a committed/expected compiler_semantics mismatch. Repro130999976438 independently refuses the same lock before a successful two-root comparison. Separate honesty success is retained, but the draft gate skips7 required groups and is not full acceptance.

The PR body's locally successful formatter rerun is useful scoped evidence, but its selected compiler/source applicability needs reconciliation with this exact-head hosted failure. These logs do not establish a wrong writer, stale executable or which digest is intended. Please resolve the VR28 reservation collision first, then bind the official lock writer to the final source/compiler and review the demonstrated affected lock/product closure before resulting-head qualification. The263-lock update by itself is not freshness evidence; no blanket rewrite or blind rerun is requested.

Independent receipt2e8a7438 retains both focused logs and the exact source predicate. Existing inspection/#439 owner retains implementation and qualification; no competing patch or runtime was launched.

@afflom afflom changed the title Inspect bounded kernel types through a typed LexLean adapter Enable owned inspection and bounded certificate validation Nov 12, 2026

@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 delta review at6a766828. The inspected UTF-8 bridge is closed to exactly the two named global helpers in language1.2, retains list/order/name checks, qualifies only those simplify references and uses a fixed simp-only/reflexivity recipe. No arbitrary tactic text, general foreign helper or proof-acceptance bypass was found in this delta. Structural missing-name construction and the budgeted null-type/empty-parameter report preserve request order; an inspection report still supplies no proof receipt. Existing base225-test/VR28 evidence is retained without rerun.

P2 — Coordinate the new SM39 allocation. spec/src/IdsSM.lex.tex:303–306 gives UTF-8 bridges SM39, but #434/#437 already reserve and implement that global ID for decimal formatting on v0.5, and #493's arrays allocation is separately contested. Absence from the #439 stack base does not make the reservation free. Please coordinate a unique disposition with those owners and carry it through the authoritative row, official projections, scenario, named test/dispatch and documentation in the existing owner wave. The earlier inspection VR28 collision with #391/#481 also remains unchanged in IdsVR; this review reserves neither replacement ID.

The delta contains269 one-line lock changes; the sampled collections lock changes only compiler identity. The reported clean906-input compiler authentication and full product closure remain owner evidence, not independently repeated here. Preserve the completed narrow proof checks and applicable products, then qualify the reconciled final head; no blanket lock/proof regeneration is requested merely for diagnosis. This remains a draft with full upstream acceptance incomplete. Existing Rule30/#439 owner retains implementation and qualification; no runtime, logs, writers or competing patch were initiated.

afflom commented Nov 13, 2026

Copy link
Copy Markdown
Member Author

Concrete namespace coordination for existing #482 ownership: reserve SM-41 for the UTF-8 bridges and VR-29 for kernel type inspection. Preserve #434/#437's earlier SM-39 decimal claim, #391/#481's VR-28 replay claim and #362's VR-24–27 range; #493 arrays is assigned SM-40 in comment6843396283. Bounded live release/queued-source and indexed reservation searches found no competing allocation of these replacements. Surface any unpublished prior reservation before applying them. Carry the mapping through authored spec/src rows, official projections, scenario/named test/dispatch and documentation in your existing batch; preserve historical receipts with their original IDs and record the rename.

The current6a766 full upstream acceptance is blocked concretely by rest job131650065410 in run43353452127:298 cases passed,13 failed. One saved focused log and independent grouping identify: GC01/GN01 calculus provenance correspondence; EX09 classify and SL01/02/05/08 ends_with package correspondence; RB07 adt-evaluation provenance (decoded JSON differs only compiler_semantics601e9f4a vs79cf6fea for that first file); NS01 noncanonical IdsPD; PK05 two additional command-driver/host-effects fixture paths; RP11 SM01–38 range vs lastSM39; TL11 commands golden manifest; and SP17 four production/preserve.rs Lean-syntax literals at1303/1314/1345/1357 (decodeAndAdd run_cmd, namespace and two import templates).

Please disposition those demonstrated source/format/register/inventory/product effects together after the namespace repair, using the existing official writer path and reusing still-valid evidence. SP17 is an actual source-audit refusal, not evidence to weaken the audit; the log alone does not establish a semantic bug or a justification for an exemption. Identity-only RB07 evidence applies to that compared file, not all products. Exact repro43353452087 and separate honesty43353452167 passed, but do not clear these failures or skipped draft groups. No competing patch, blanket regeneration, rerun or stopped431/TC03 experiment was started. Log SHA2560c7bbf26aa213a9ebb0e2ef2542f23ddcb38a0d7ca1140f13565cbac2ccea74c is retained; no repeat log download needed.

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