Skip to content

Integrate certificate fixes, GNAF request validation and nested induction - #393

Merged
afflom merged 13 commits into
release/v0.4from
claude/issue-314-310-kernel-rejections
Nov 5, 2026
Merged

afflom merged 13 commits into
release/v0.4from
claude/issue-314-310-kernel-rejections

Conversation

@humuhumu33

@humuhumu33 humuhumu33 commented Nov 2, 2026 •

Copy link
Copy Markdown
Contributor

Closes #314.
Closes #310.
Closes #432.
Closes #429.

This carrier fixes the recursive-argument and binder-arm certificate failures, adds certified BS16 GNAF request validation and SM37 nested induction, and integrates them with accepted Host/DF19. It preserves the original #393/#390/#292 histories and existing schema, runtime and axiom-policy guarantees.

Official identity and product preparation completed once, with independent source, artifact and semantic reviews: renderer/semantics, model/spec/format, calculus, valid-lock fixed point, protected negative inputs, all-five exports, goldens, 309 fixtures and normalized verification for all 22 non-Atlas projects. Existing observed axiom sets and negative codes remain intact.

The initial hosted gate exposed a strict BS06 census mismatch and two noncanonical AU04 sources. The reviewed follow-up calibrates the strict expectation to the measured 583 without changing its factory, bounds, seeds, differentials or plants, and adopts the official formatter's semantic-preserving outputs. These repaired cases and their complete partitions passed on the final head.

Final head 050e96579b83f2f71dda78e1bf6a87791278582a, tree 3065516167862533e0594427c28be8fe74406aae: CI 41923290770 passed all 45 jobs, including honesty, every normative partition, supported hosts and Atlas; reproducibility 41923226337 also passed. All 49 SM37 and 21 BS15 plants and strict counter obligations remain part of the accepted gate.

#390 and #292 are superseded deliveries through this carrier. Actual merged-tree equality is checked after landing before explicit scoped issue closure on release/v0.4. #278 has a separate full-scope reassessment; #275 and broader unfinished work remain open. #391 and the v0.5 wave are outside this carrier.

afflom and others added 5 commits October 22, 2026 20:43
…ion (#278, R3: row, scenario, named test, first negative fixture)

R3: the register row, the SPEC.md section 31 row, the scenario, the failing
named test conformance_sm_37, and the negative fixture induction-nested-type,
which records that `induction` on a type that holds a list of itself is
refused as LLT4001 with a message that names `nested_induction`. No
implementation: the fixture fails until the message changes. The count moves
446 to 447. Stacked on the DF-19 branch.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01F4DJnZorC2Jd4MJfN9DbZk
… through its recursor (SM-37, #278)

A language-1.2 theorem about a value of a nested inductive type (one that holds
itself under List, Option, Result or a product, in no mutual group) may be
proved by nested_induction: the proof names a parameter of the theorem, states
one proposition for each auxiliary member of the type's recursive family (the
family DF-19 computes) and gives a branch for every constructor of the type and
of each member. The linker checks the members, the branch sets and the binder
counts level by level before any backend; the backend lowers the step to
`induction x using T.rec (motive_n := fun b => P)` and one positional
`next binders =>` per minor premise, in the recursor's order whatever order the
branches are written in. `induction` keeps refusing a nested type, and its
message now names nested_induction.

The theorem's scrutinee must be a theorem parameter so the backend knows its
type to name T.rec; Environment.proof_parameters carries them. Both schemas gain
the step. language/semantics-1.2.toml: semantic_ir 15 -> 16, proof_lowering
3 -> 4, which moves the 21.2 identity; the locks, records, goldens, fixtures and
packages the digest binds are not regenerated here.

examples/nested-induction proves leaves_le_size over Json by nested_induction,
verified under Lean 4.32.1 with exactly the axioms propext and Quot.sound, which
the SM-37 row now records. conformance_sm_37 refuses 49 plants as LLT4001
(reordered members, wrong binder counts and a missing branch on each level, a member of another instantiation, ...). A Rose (Rose Nat) case pins that the family counts a group only at its own type arguments.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01F4DJnZorC2Jd4MJfN9DbZk
…, Option and Result members (SM-37, #278)

- examples/nested-induction carries no lexlean.lock: every lock is
  regenerated at landing on the identity current then. conformance_sm_37 locks
  its own copy of the example.
- backend/semantic.rs: a nested_induction scrutinee with no named parameter is
  an LLI9001 diagnostic (proof and proof_branch return Result), not a guessed
  `<local>.rec`.
- conformance_sm_37: Lean-backed positive cases with one member under Option
  and one under Result, with the goals asserted in the recursor's order, so the
  none,some and error,ok rows of family_constructors are exercised; the
  language-1.1 use and the four scrutinee misuses go through refuse_project
  (LLT4001, the message, no backend output).
- The SM-37 row, SPEC section 31 and the scenario say the proof names a
  parameter of the theorem; CONFORMANCE.md regenerated.
- VERIFICATION.md: the lib-test count is 213.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01F4DJnZorC2Jd4MJfN9DbZk
Certificate A reduces a structural definition on its declared argument, but the
Lean backend stated the argument only for a mutual group. Lean took the first
parameter every recursive call decreases, so `paramsOf names types ...` recursed
on `names`, stuck on the free list at `Ty.TUnit`, and the kernel refused
`__rel_N` (LLV7013).

The backend now states the argument when a definition is executable and matches
a parameter before it. No committed definition has that shape. `lean_backend`
moves with the regeneration of this branch, as I10 and §30.1 require.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
`fold` read only a returned move or copy and runs before the last-use moves, so
`let x = (*b).clone(); x.clone()` was left alone and became
`let x = (*b).clone(); x`. Certificate B's `mCons` states an arm through `foldB`,
which folds that block, so the kernel refused the arm the crate wrote (LLV7015),
and no certificate could state it.

`fold` and `returnedLet` of `RustBodies` now read a returned clone too. `lower.rs`
is a file of the renderer, so `rust_renderer` and the language-1.2 identity move
with the regeneration of this branch.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
humuhumu33 and others added 2 commits November 2, 2026 21:43
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Draft #295 (#116, the overflow bound) names SP-31 in the same preservation
suite for a different claim. Its branch is not merged, so the two would collide
at the next forward merge of this line.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@humuhumu33 humuhumu33 changed the title Fix the certificate rejections of #314 and #310 (SP-31, SP-32) Fix the certificate rejections of #314 and #310 (SP-32, SP-33) Nov 2, 2026
The gate reads a row of four cells, so these two were not counted (RP-07: 468
rows, not 470).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

afflom commented Nov 3, 2026

Copy link
Copy Markdown
Member

Maintenance review of 7184a94: the source changes address the two reported certificate rejections; no additional demonstrated functional defect emerged in review. The current draft CI stops at calculus::rust::tests::the_compiler_semantics_records_the_renderer (218 passed, one failed), so it supplies no SP-32/SP-33 or full-gate evidence yet.

Landing order remains #389, then this PR. After updating to that integrated release/v0.4 head, please commit the lean_backend/rust_renderer identity changes and the official regeneration, reconcile the register counts, and run SP-32/SP-33 plus the affected BS-11/13/14 differential cases before marking ready for the complete gate. Preserve the recursive/non-executable controls and boxed-field/overflow fixtures already added. Fixes #314 and Fixes #310 are the right issue scope; no expansion is needed.

Not for the first push: this commit is to be dropped if the regeneration wave
does not prove it out.

keepE and keepP, the identity functions that handed a binder of the matched type
through an arm, are gone from RustBodies: their eleven call sites are the
binders themselves. paramsOf of RustAssemble takes the list of names before its
recursive argument again, and the backend now states that argument. The note on
CERTIFICATE_BUDGET and the paragraph of SPEC.md section 17.23 that named both
workarounds go with them.

Only the source is edited. `lexlean check` of a copy of bootstrap and stdlib
accepts the 19 modules; no certificate has been checked, and the committed
expected outputs, production packages, counts and records still describe the
old source.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

afflom commented Nov 3, 2026

Copy link
Copy Markdown
Member

Latest-head refresh: eee82cd removes keepE/keepP and restores paramsOf ordering in authored bootstrap source. The commit explicitly reports no new certificate checks, and the generated outputs remain unchanged. Please refresh the PR body's 'still in the tree' workaround paragraph when updating the branch so reviewers do not mistake the older description for this head.

These source removals belong in the already-planned certificate-fix identity wave after #389; do not run a separate workaround-removal regeneration. The prior source approval and old SP-32/SP-33 measurements do not establish this new source's acceptance. Final integration still needs the reviewed pins, official writers, SP-32/SP-33 and affected BS-11/13/14 evidence, then complete CI. No additional patch or unchanged-head rerun is requested.

afflom commented Nov 4, 2026

Copy link
Copy Markdown
Member

Next-wave source planning has been refreshed against the settled Host lowering fixes, without changing the active Host checkout or starting another writer. Current eee82cdf applies textually except the three documentation/history overlaps; its returned-Clone fold change composes with the new thin Build dispatcher. Preserve both dispatchers and the restored bootstrap parameter order in the final source review. This rehearsal is not integration acceptance.

Settle the reviewed BS16 source from #390 before this next freeze so the same required bootstrap export/golden/verification pass can carry it. The existing follow-up 30171790ee99469242723e7cf3057bdbd514df0a contains the Unicode and BS15 prose corrections still omitted by #390's current ffb352 head. Its source overlap with this wave is confined to README/history; no duplicate implementation branch is needed. Preserve the real PR ancestry and scoped closures (#314/#310 here, #432 for BS16; #275 remains open).

After #389's final acceptance, refresh release and contributor heads, settle source and registered claims, then compute the final renderer and lean_backend identity changes once. Preserve language1.0/1.1 and unchanged runtime fingerprints. The demonstrated compiler-manifest dependency requires calculus package metadata to be refreshed after golden generation; final comparison must use that actual manifest. Old corpus tuples are not final combined evidence: measure the merged corpus while keeping strict assertions, differential samples and plants. SP32/33 and affected BS evidence plus the complete final-head gate remain required. No pins, writer, test, CI or branch mutation ran for this planning audit.

Preserve the original contributor and reviewed repair histories while keeping
BS16 pending beyond the scoped Host integration. Retain current Host fixes,
Unicode corrections, strict corpus mirrors and CI implementation.

This is a source-only draft restack. Model projections and generated bootstrap
artifacts remain pending in the coordinated next wave. Skip this knowingly
incomplete draft run; complete changed-head CI is required before readiness
and merge. No validation or acceptance is claimed by this commit.

Closes #432

afflom commented Nov 4, 2026

Copy link
Copy Markdown
Member

The coordinated next-wave source now has concrete published ancestors: Host/#402 integration88b078cc529169c53f4eb2e657781ebd96d2527c is ready in #389 and awaiting fullCI41781980948; existing draft #390 explicitly reapplies BS16 at7ca5a7ff369135c1c700cb376c19cc18ba227489 with parents88b078+30171790. This retains the reviewed Unicode/count corrections and current Host/CI/profile fixes without letting retained old390 ancestry mark BS16 delivered. The25-path redelivery is source-only: official projections, bootstrap artifacts and combined corpus measurements remain pending, not validated by inherited records.

Continue the already-reviewed #393/#390 source reconciliation after Host acceptance, then one coordinated identity/artifact wave; do not regenerate independently on the old bases. The existing initial calculus writer and final post-golden refresh/comparison dependency remains documented in the source plan. #314/#310/#432 remain open until their accepted delivery; broad #275 stays open.

afflom commented Nov 4, 2026

Copy link
Copy Markdown
Member

Post-Host source-freeze update: #284’s DF-19 is now accepted on release/v0.4, and Host/#402 at943bba379a is in its complete combined-head gate (41802893306). The accepted SM-37 source in existing #292 has also been reconciled against that provisional Host tree and independently source-reviewed: original implementation/49 plants retained, strict schema union preserves Host definitions, only the intended semantic_ir16/proof_lowering4 axes advance. #292 remains draft, now correctly targeting release/v0.4; its official artifacts and acceptance are still pending.

Include that reviewed SM-37 source in the next source-union review alongside this certificate fix and #390’s explicit BS16 redelivery, before computing identities or running writers. This can share the already-required compiler/fixture/export/verification regeneration instead of scheduling another identity pass solely for SM-37. Reconcile the actual backend/schema/count union and real parent histories first; do not copy stale pins or assert a guessed combined count. #292’s preparation is source evidence only, not permission to skip its SM-37/DF-13/DF-19 cases or final gate.

No new wave starts before Host acceptance. Preserve the established post-golden calculus dependency and one final comparison, all strict corpus/plants, and the scoped issue boundaries (#314/#310 here, #432 for BS16, #429 for SM-37). Broad parents stay open until their complete intended work is actually accepted.

afflom commented Nov 4, 2026

Copy link
Copy Markdown
Member

The Host prerequisite is now accepted: #389 merged into release/v0.4 as c613290 after all44 final jobs plus honesty/repro passed, with actual/tested/reviewed tree equality. #390 remains open and draft outside Host ancestry.

I am advancing the already-coordinated source union of this exact eee82cd certificate fix, #390's reviewed7ca5a7ff redelivery, and #292's accepted SM37 preparation against that actual release landing. Existing contributor heads are unchanged. This step is source reconciliation and independent review only; no guessed combined count, copied stale pin, writer or new matrix. Preserve the Host/DF19 fixes, strict schemas/corpus comparisons and every scoped acceptance case. The established initial-calculus/post-golden dependency and single final comparison plan remain in force for the later admitted wave.

Scoped #314/#310/#432/#429 remain open until complete accepted delivery. Broad parents remain open. #391's separate measurement work is not silently added to this source freeze.

afflom commented Nov 4, 2026

Copy link
Copy Markdown
Member

The combined source review is clear against accepted Host tree65aa81df: this certificate fix, #390's explicit redelivery and #292's accepted SM37 implementation compose as source tree0328ab561af4adbe2a17cdf6489b792b74133a6e. The 70-path delta retains Host/DF19 behavior, the strict schemas and corpus comparisons, Unicode repairs and the complete 599-line SM37 case with 49 plants. Static registration is 475 matching claims/scenarios/tests and 303 prescribed negatives; these are not measured generated-corpus totals.

The combined identity assessment includes both changed v2 schemas, not just the older certificate-only source. Proposed axes are lean_backend22/semantic_ir16/proof_lowering4; existing fingerprint/product values remain provisional. The first execution plan uses the official renderer calculation and new correctly rooted frozen executables before any downstream writers. Its concrete workspace/resource plan is still being reviewed; no identity calculation, build, writer or new gate has started. #391's separate controlled measurement currently owns the local phase lock and is not included in this v0.4.2 union.

Scoped #314/#310/#432/#429 remain open pending actual runtime/artifact and final-head acceptance. Actual release and contributor ancestry will be retained; no audit-tree history will be published.

afflom commented Nov 4, 2026

Copy link
Copy Markdown
Member

The concrete first identity-phase driver is independently reviewed, and the separate #391 measurement has completed and released the local phase lock. Current contributor heads remain eee82cd / #3907ca5a7ff / #2920b0c95d; live release/v0.4 is actual c613290.

I am proceeding with the bounded first phase on source union0328: reuse the existing clean worktree while retaining its old Git reference, materialize the actual accepted release plus reviewed source, compute/check the renderer record through the official exact unit, then build and freeze both executables with actual selected-root/input and all-language identity observations. It stops before any product writers. Hardlinked Cargo executables must become independent frozen copies; merely renaming one alias is insufficient. The reviewed driver also guards source before changing the pin and compares the complete embedded path set.

Two completed, clean checkouts were retired using ordinary Git worktree removal, with their source references, restore commands and external evidence retained. This recovered 986MB; it is not a capacity guarantee. Any unexpected failure stops with source and logs preserved. No full gate, accepted Host replay, or concurrent regeneration is being repeated. The official downstream wave and final-head gate still require their actual outputs and acceptance.

afflom commented Nov 4, 2026 •

Copy link
Copy Markdown
Member

The next coordinated #393/#390/#292 wave has completed its bounded identity and input phases. The official renderer is d75eb77f… and language 1.2 identity is 635a25ac…; language 1.0/1.1 identities are unchanged. Correct-root frozen executables and all 905 embedded inputs are recorded. Initial model/spec/fmt and whole calculus writer passed; the 166 calculus changes are compiler-identity provenance only.

The subsequent official lock phase passed all 49 commands: 24 initial relocks, a 24-lock byte-stable fixed-point pass, and the new nested-type negative lock. Independent review confirms the exact candidate delta: 19 publishable valid locks, one new negative lock, 239 compiler-identity field substitutions and three request target-digest substitutions. Other negative bytes/codes, older/exceptional locks and deliberately absent locks remain preserved. Five local-only locks and 24 newly owned empty guard directories were archived outside the candidate.

These are producer/input results, not runtime or full-gate acceptance. The single all-five production-export writer started at 15:14:52 UTC after independent driver review and a fresh head/ownership refresh. Entry checks confirmed all 905 embedded inputs, errors and frozen executable hashes unchanged. It owns the whole local phase lock and stops at its terminal result; golden, fixture and verification writers have not started. Existing contributor heads remain unchanged, and scoped #314/#310/#432/#429 remain open. No completed Host gate or focused case was repeated.

afflom commented Nov 4, 2026 •

Copy link
Copy Markdown
Member

The single all-five production-export writer passed in 3013.252s. Independent artifact review reconciled all 243 changed paths and 118 actual packages: 31 bootstrap, 3 helpdesk, 10 production, 52 production-coverage and 22 stdlib. The review explicitly retained five genuine bootstrap program/Rust changes from RustBodies/RustAssemble propagation and the new BS16 validate_request package; these were not hidden by identity normalization. Other changes are source/compiler provenance. Source, locks, negative inputs/codes, helper ownership and all 905 frozen inputs remained intact.

The conditionally admitted golden sequence also passed: official golden write (58.475s), required post-golden calculus write (22.667s), and one final comparison (10.623s). Independent review cleared the actual 132 golden and 351 calculus changes; the final comparison changed zero files, with 977 calculus files equal. Atlas stayed unchanged. New BS16/SM37 outputs and existing RustBodies/RustAssemble changes are classified explicitly; post-golden calculus changes are source provenance only. No binary rebuild or completed initial writer was repeated.

One official whole fixture writer started at 16:58:29 UTC after independent driver review and fresh head/ownership checks. It covers the 309 discovered cases and must retain every prescribed exit/code and negative input; the new nested-type refusal is not claimed passed before execution. The verification-writer continuation is source-reviewed and conditional on actual fixture completion plus independent artifact/code review, not elapsed time. Verification records, measured combined corpus checks, scoped focused acceptance and complete final-head CI still remain. Contributor heads and all scoped issues remain unchanged/open; this is producer evidence, not merge acceptance.

afflom commented Nov 4, 2026

Copy link
Copy Markdown
Member

The official fixture writer passed once at 17:16:17 UTC in 1065.081s. Independent actual review cleared all 309 unique cases: every existing code/exit and input stayed intact, the new nested-type case emitted LLT4001/exit 1 with no artifacts, and the exact 16 changed expected files comprise 12 contextual identity updates plus four new outputs. No hand-edited oracle or repeat writer was used.

After that review and a fresh unchanged release/contributor-head check, the already conditionally authorized verification writer started once at 17:19:01 UTC. It covers the actual 22 selected projects with the supported without:uor-atlas selector, including nested-induction. Entry checks confirmed unchanged 905 embedded inputs, errors and both frozen executable hashes. It is still active; terminal guards and independent record review remain pending. No focused tests, Atlas, full CI or publication have started. All scoped issues remain open.

afflom commented Nov 4, 2026

Copy link
Copy Markdown
Member

The single 22-project non-Atlas verification writer completed at 18:24:23 UTC in 3918.668s. Independent review cleared the actual 2,260 normalized records and 1,396 successful process records; all 2,998 changed paths are selected verification outputs. Source, locks, negative inputs/codes, Atlas, the 905 embedded inputs and both frozen executables stayed intact.

A separate semantic review found no changed observed axiom set for existing declarations. The five changed bootstrap programs match the reviewed RustBodies/RustAssemble changes, and the new BS16/nested-induction records are classified explicitly. The existing R25 policy admission is preserved.

The authorized read-only corpus ledger then completed once. Its 284 package files, 122 fixtures and 201 JSON program keys match the corresponding source expectations; the six added/five removed equivalence classes are exactly the five replaced bootstrap programs plus validate_request. This is JSON membership evidence, not a claim that lowered/vector/refusal/plant counters passed. No numeric assertions changed.

I am preparing the existing PR's final combined tree and honest verification record for independent review. One final hosted full gate will own all affected cases and unchanged obligations; there is no extra broad local gate. #390/#292 and scoped issues remain open until the reviewed combined acceptance and closure sequence. No commit has been published yet.

Retain reviewed contributor histories and accepted release/v0.4.
Official identity and artifact wave complete; full final-head hosted acceptance pending.

Refs #314, #310, #432, #429.
@afflom afflom changed the title Fix the certificate rejections of #314 and #310 (SP-32, SP-33) Integrate certificate fixes, GNAF request validation and nested induction Nov 4, 2026

@afflom afflom left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

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

Reviewed final source/artifact tree 96aa424e1da1b2276f36029f8d302c1e60d3f683 at 3fe13b5. Exact ordered parents retain original393 eee82, actual accepted release c613, reviewed3907ca5 and2920b0c. Independent review matched all10,639 final paths (10,631 file hashes/eight symlink targets) to the actual terminal writer inventory; only the honest verification appendix differs from the producer tree. No audit/disposable commit is a parent.

Official identity/model/calculus/locks/export/golden/fixtures/non-Atlas verification results and independent semantic/corpus reviews are preserved. Negative outcomes and existing observed axioms remain intact. JSON membership is not a typed/lowered/vector/plant pass; all strict obligations remain for the final hosted gate. No duplicate local heavy cases or writers were run.

Source and artifact readiness support requesting the complete final-head hosted CI, including all changed SP/SM/BS cases, unchanged cases, supported hosts, Atlas, same-run honesty and reproducibility. This COMMENT is not merge acceptance or self-approval. Do not merge until the entire final gate passes and final head/base/reviews/tested tree are rechecked. After combined acceptance, explicitly supersede390/292 before carrier merge; verify actual merged-tree equality afterward and close only314/310/432/429. Broad parents remain open.

@afflom
afflom marked this pull request as ready for review November 4, 2026 18:39
@afflom
afflom marked this pull request as draft November 4, 2026 19:33

afflom commented Nov 4, 2026

Copy link
Copy Markdown
Member

The final-head gate at 3fe13b50 has demonstrated two failures, so I restored this PR to draft. The current run is left intact to collect its remaining evidence; no rerun or cancellation is requested.

  • BS-06, job126784187495: the strict vector tuple is (583,2145,400,394,2989) versus expected (582,2145,400,394,2989). Earlier counters in that execution report 284 manifests, 122 fixtures, 201 programs, 260 crates, 284 packages; 687 printCrate vectors (520 committed distinct/65 refused) and 2145 printExpr vectors (1532 hand-built/0 refused). The JSON-only local ledger did not establish this private vector count. A bounded causal review must identify the additional counted vector before any assertion/mirror correction; strict comparisons remain.
  • AU-04, job126784187554: tests/fixtures/binder-arm/src/Main.lex.tex and tests/fixtures/recursive-argument/src/Main.lex.tex are not their own canonical source. That leg reports 90 passed/1 failed. This is an authoring-source failure, not evidence that a verification case failed. A separate bounded repair review will use the official formatter and preserve fixture semantics, helper ownership and all acceptance obligations.

Same-run honesty and reproducibility passed, but do not accept this failing head. All scoped issues and #390/#292 remain open. Existing official producers and completed cases will not be repeated speculatively; any changed-head final gate follows reviewed repairs. No merge or closure is authorized by partial green jobs.

Keep the existing typed factory and 6000-node predicate strict while updating
its observed evaluator census and all normative mirrors from 582 to 583.
Adopt the official formatter's two fixture-source outputs and regenerate the
conformance projection through the official model writer. Preserve failed CI
and published verification history; no individual crossing cause or later
BS06/AU04 case pass is claimed.

Validation: official model write, validate-model, validate-spec-links, cargo
fmt --all -- --check; independently reviewed eight-path actual delta.
Complete final changed-head CI remains required before merge.

afflom commented Nov 4, 2026

Copy link
Copy Markdown
Member

The two known gate causes now have an independently reviewed local follow-up, 050e965 (tree3065516167862533e0594427c28be8fe74406aae, single parent3fe13b50). It has not been pushed; the current hosted run remains untouched until terminal evidence is collected.

BS-06 retains the same typed factory, 6,000-node membership predicate, seeds, refusals, differentials and 13 plants. The strict expected census and all normative mirrors now state the actual measured583. Contract review confirmed that identifying an individual historical crossing is explanatory work, not an additional acceptance requirement; no such cause is claimed and the failed case had not run its later legs.

For AU-04, one official frozen-CLI lock/format sequence on each external fixture copy produced only declaration field ordering and two match-layout changes. The formatter's existing semantic-ID and canonical-fixed-point checks passed; no helper, lock or oracle was copied into the candidate. Independent review cleared those actual emitted bytes.

After source adoption, one official model writer changed only CONFORMANCE. Model validation (475 IDs/86 codes), spec links (475 rows) and formatting passed. Independent actual-delta/final-tree reviews cleared eight paths, with unchanged905embedded inputs, errors, frozen binaries, production artifacts and published verification history. No identity/product regeneration or extra heavy local case was run. This is repair preparation, not BS06/AU04 runtime acceptance: the complete changed-head CI remains mandatory. #390/#292 and all scoped issues stay open.

@afflom afflom left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

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

Reviewed repaired head050e96579b83f2f71dda78e1bf6a87791278582a, tree3065516167862533e0594427c28be8fe74406aae, single actual3fe13b50 parent. The independent final-tree review2b4d81a0 and actual-delta review67b0a369 bind the eight changed paths and preserve original contributor history, all identity/product bytes and prior verification records. The only final-capture addition is the honest completed repair-phase record.

OldCI41889333456 is now terminal:39successful/6failed jobs. Every failure—including final Intel473pass/2fail—reports only the reviewed BS06 strict census583vs582 and AU04 canonical-source defects in binder-arm/recursive-argument. Package and all Atlas jobs passed on that prior tree. These old-head passes are not acceptance of this follow-up.

The strict census and authoritative mirrors are calibrated to the measured583 with factory/nodebound/seeds/differentials/plants unchanged. The two actual official formatter outputs retain semantic-ID and canonical fixed-point guarantees. One official model projection and model475IDs86codes/spec475/fmt passed; completed producer and heavy local cases were not repeated. No repaired BS06/AU04 runtime pass or precise historical vector-crossing cause is claimed.

Source/artifact readiness supports requesting ONE complete changed-head hosted gate. This COMMENT is not self-approval or merge acceptance. All required cases/plants/hosts/Atlas, same-run honesty and reproducibility must pass, followed by fresh final head/base/review/tested-tree checks. Keep390/292 and scoped issues open until the previously recorded supersession/accepted-landing sequence.

afflom commented Nov 4, 2026

Copy link
Copy Markdown
Member

The repaired cases now have exact-head hosted evidence at050e9657, run41923290770, syntheticfb159f03/tree30655161:

  • AU04 passed by name at22:19:09; verification partition91passed/0failed/384filtered,1290.16s (job126902756989).
  • BS06 passed by name at22:59:35; its full bootstrap-print partition1passed/0failed/474filtered,2470.43s (job126902757081). This execution reached the later obligations that the old strict-census failure had prevented.

Each completed job log was fetched once and preserved; no local case or writer was repeated. Receipt hashes d832d5bf (AU04) and183177eb (BS06), full log hashes a94ae172 andf8ad29dd, bind the results. This establishes these repaired cases, not complete merge acceptance. At23:34,41 of45 materialized jobs had passed; Windows, Intel, bootstrap-package and bootstrap-check were still running. Honesty/reproducibility passed. Keep the current run intact and all scoped issues/#390/#292 open until the complete final acceptance and recorded supersession/landing sequence.

@afflom afflom left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

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

Final acceptance reviewed at head 050e965, tree 3065516167862533e0594427c28be8fe74406aae. Complete final CI 41923290770 is terminal SUCCESS: all 45 jobs passed, including same-run honesty, required hosts, every normative gate partition and the complete Atlas shard/join sequence. Exact-head reproducibility 41923226337 passed. Repaired AU04/BS06 named and complete partition results are preserved; their later obligations were executed. No skipped/incomplete/failing required job is being counted as acceptance.

Independent source/artifact/history reviews and terminal gate review retain the actual official writer evidence, strict counters, negative outcomes, observed axiom policies, 49 SM37 plants, 21 BS15 plants and contributor ancestry. No completed producer, heavy local case, measurement or gate was repeated. The historical failed head and repair evidence remain intact.

Fresh head/base/reviews/ownership are unchanged; no unresolved review request is present. Tested synthetic fb159f0 has parents c613290 + 050e965 and exactly the reviewed tree. This COMMENT records maintainer evidence review, not self-approval. Explicitly supersede drafts #390 and #292 before this carrier merges, then verify actual merged-tree equality and close scoped #314/#310/#432/#429. #278 has a separate full-body reassessment; #275 and broader unfinished work remain open.

@afflom
afflom merged commit 55be551 into release/v0.4 Nov 5, 2026
46 of 56 checks passed
afflom pushed a commit that referenced this pull request Nov 5, 2026
Retain the contributor lexer changes and actual #393 landing. Refresh the
bootstrap products with the official export, golden and verification writers.
Record independently reviewed producer evidence; final combined CI is pending.
humuhumu33 added a commit that referenced this pull request Nov 5, 2026
The line took #393 (SM-37, SP-32, SP-33, BS-16) with a compiler-semantics move, so
every lock, record, golden, fixture and package was regenerated there. This merge
takes the line's oracles, records, export and bootstrap counts as they are, keeps
the SL-11 row, module, scenario and changelog bullet, and sets the register to
476. The Json packages' export, stack bounds, build oracles, verify records and
count mirrors regenerate on this tree in the commits that follow.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
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.

2 participants