Skip to content

SM-39: arrays of fixed-width integers in language 1.2 - #493

Merged
humuhumu33 merged 33 commits into
release/v0.4from
feat/verified-arrays
Nov 16, 2026
Merged

humuhumu33 merged 33 commits into
release/v0.4from
feat/verified-arrays

Conversation

@humuhumu33

@humuhumu33 humuhumu33 commented Nov 12, 2026 •

Copy link
Copy Markdown
Contributor

Arrays of fixed-width integers for language 1.2 (SM-39, SPEC.md §17.12, Arrays). Gap A of three on the way to a verified inference engine written as a LexLean application; gap B (contracts for external code) is the proposal #492 and gap C (floating point) is the proposal #486. The design, the evidence and the decisions left to the owners are in docs/proposals/verified-arrays.md, the first commit.

Why

An inference engine reads and writes sequences of machine integers by position. Language 1.2 has lists (linear index), byte strings (no write) and no proof story for a loop over either by index.

What

  • Type. array{element: T} over one of the eight fixed-width integer types, lowered to the pinned Lean's Array T. Any other element is LLT4001.
  • Operations. length and index now also apply to an array. Three new primitives: array_from_list, array_to_list, array_set. A write is total: at or past the length it leaves the array as it is.
  • Runtime. A sixth native module, LexLeanRuntime.Arrays (15 declarations, exported once from authored Lean and checked to regenerate to the same environment, tests/runtime-library/ARRAYS.md). It holds the two instances, the three primitives, and nine laws that relate an array to the list of its elements. Every row is within propext.
  • Proofs. No new proof form. A law proof states an array law as a document theorem; simplify rewrites with it. The law registry gains an optional module per row, and a law proof imports the module that states its law, so a module that applies only array laws does not import LexLeanRuntime.Laws (which does not terminate under C output, LexLeanRuntime.Laws loops forever when Lean 4.32.1 emits C for it (breaks Prism app builds, PP5007) #454). The arrays module compiles with C output in 15.7 s.
  • Production. The array type and the three primitives are formal-only in language/production-1.2.toml, as text components are; a root that reaches one is LLT4005.
  • Execution. Constant-time indexing and the in-place write of a singly referenced array are the pinned Lean's, cited as LEAN-ARRAY-4-32-1 in model/authorities.toml and not claimed here.

No existing language-1.2 document changes meaning or is refused, and no generated Lean or LaTeX module of an existing project changes: the regenerated goldens differ only in build IDs, declaration IDs and source maps. Languages 1.0 and 1.1 are untouched.

Acceptance

examples/arrays defines an int8 dot product over two arrays by index and proves, for all arrays,

dot left right = dotList (array_to_list left) (array_to_list right)

where dotList is the plain recursion on two lists. It also rectifies an array in place and proves the length is kept.

lexlean verify in examples/arrays (pinned Lean 4.32.1):

verified 1 module; attestation edd3c119ebfd7583b2d9a85e4015f0bae6d473929b5a57d11be800f4a9da116b

(a Windows host run). The committed Linux x86-64 records, written by the regenerate workflow, state the same in examples/arrays/expected/verify/audit/output.txt:

'Arrays.Main.dot_equals_list' depends on axioms: [propext, Quot.sound]
'Arrays.Main.rectify_keeps_length' depends on axioms: [propext]

Observed axioms, equal to the declared sets for every declaration:

Declaration Observed axioms
dot_equals_list Quot.sound, propext
loop_reads_list, shift, index_sums Quot.sound, propext
array_index, rectify_keeps_length, set_keeps_length propext
array_length, the five list-indexing equations none

Quot.sound comes from to_int and the linear_arithmetic lemmas, not from arrays. Nothing depends on Classical.choice.

Tests

  • conformance_sm_39: the runtime module's shape and axioms, the generated Lean and document of the example, the import rule, 17 linking refusals, the language-1.1 refusal, the production refusal, and with the pinned Lean the example, a round trip at each of the eight element types, a loop that advances by two refused with LLV7002, and a law applied to a statement it does not prove refused with LLV7002. It passes locally.
  • Five negative fixtures: array-element-not-fixed-width, array-set-wrong-element, array-pattern-match, array-under-1.1 (LLT4001), array-production-root (LLT4005).
  • Four planted defects, each refused by the case (VERIFICATION.md): the element check accepting every type, the arrays import dropped, every law proof importing the laws module, and array_set typed to another element.

Identity

semantic_ir 16 to 17 and lean_backend 22 to 23; the language-1.2 compiler-semantics ID moves. The locks and records that name it, the goldens, the fixture expectations, the calculus fixtures, the production packages, and the verification records of the language-1.2 projects are regenerated by the repository's writers in the chore commits: the goldens by check-golden --write, and the rest by the regenerate workflow (run 43241009907, Linux x86-64, 4,390 paths; no bootstrap count moved). examples/uor-atlas is a language-1.0 project and is untouched. PR #474 moves the same identity; whichever lands second regenerates once more.

Decisions for the owners

In the proposal: total array_set against an Option result like #443's update; element types; production as formal-only; a readable (Array T) notation; the register number (SM-39 is also used by the unlanded #434 on the v0.5 line); packed storage.

🤖 Generated with Claude Code

humuhumu33 and others added 9 commits November 12, 2026 12:14
The design, the evidence, and the decisions left to the owners for the language-1.2 array type: one type over the eight fixed-width integers, lowered to the pinned Lean's Array, with length and index extended to it, three primitives, and nine laws that relate an array to the list of its elements.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
… and section, scenario, and the failing named test

The register is 494 IDs. conformance_sm_39 is named and no case is wired for it, so it fails until the implementation lands.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
LexLeanRuntime.Arrays, exported once from the authored Lean in tests/runtime-library/arrays-oracle/: two instances, three definitions, and ten theorems, every row within propext. The regenerated environment equals the export in all 15 declarations (tests/runtime-library/ARRAYS.md).

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The type array{element: T} over one fixed-width integer type, lowered to the pinned Lean's Array; length and index on an array; the primitives array_from_list, array_to_list, and array_set; nine array laws in the registry, each a declaration of LexLeanRuntime.Arrays, with a law proof importing the module that states its law; the array constructs formal-only for production; schemas, the oracle overlap table at version 5, the authority LEAN-ARRAY-4-32-1, examples/arrays, five negative fixtures, and conformance_sm_39. semantic_ir 16 to 17 and lean_backend 22 to 23: the language-1.2 identity moves.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…that names it

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…s edited in the notation; CONFORMANCE.md follows the register

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…e-1.2 identity

cargo xtask check-golden --write. No generated Lean or LaTeX module of an existing project changes; the changes are build IDs, declaration IDs, and source maps, and the new examples/arrays/expected/build.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…CATION.md records the planted defects

The audit of SPEC.md 17.17 refuses a rest pattern in production/lower.rs, so the array type is written with its field named in every production match. VERIFICATION.md states what conformance_sm_39 checks and the four planted defects it refused.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

@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.

Bounded source review at b105f8f:

  1. P2 — Coordinate the SM-39 identity before forward integration. The live #434/#437 lane already reserves and implements SM-39 for decimal formatting on v0.5 (current #437 head77f6dd2). Arrays adds a different statement and the same conformance_sm_39 test on v0.4. There is no same-tree duplicate yet, but the release branches must meet. Proposal D5 must not presume renumbering the existing contributor's claim. Agree the disposition with that owner and carry it consistently through the register, scenario, test and related documentation/projections. Preserve valid proof receipts and incorporate only necessary derived changes into the coordinated writer work.

  2. P2 — Qualify allocation/in-place claims (verified-arrays.md D1 and execution discussion). “A total write allocates nothing” and threading an array through a loop do not establish runtime exclusivity. The document itself correctly says a shared array is copied. Describe avoiding an Option result separately, and keep in-place mutation conditional on the pinned runtime's exclusivity check. The value/length theorems are not ownership or allocation evidence.

The bounded review found matching row/scenario/test bindings and explicit formal-only production refusal; it is not a full audit of the regenerated product set. Please link the preserved initial pre-implementation named-failure receipt when assembling acceptance evidence—do not recreate it. The draft body correctly retains incomplete products and verification records; exact-head CI43222211167 failed, repro43222211201 passed. No CI failure cause is inferred from metadata. Existing owner retains implementation/qualification; no runtime, writer or experiment was repeated.

humuhumu33 and others added 5 commits November 12, 2026 14:45
…es; the README lists examples/arrays

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
… samples the array type

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
… production packages, and verification records for the new language-1.2 identity

The patch of the regenerate workflow, run 43241009907 on regenerate/verified-arrays at 8d9ba27 (Linux x86-64, pinned Lean 4.32.1): 4390 paths. No count of the bootstrap count cases moved. examples/uor-atlas is a language-1.0 project and is untouched.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
@humuhumu33
humuhumu33 marked this pull request as ready for review November 12, 2026 18:50

afflom commented Nov 12, 2026

Copy link
Copy Markdown
Member

Bounded readiness readback at859b52ca: the prior missing-product status is superseded. Complete non-truncated Git-tree inventories show regenerated array fixture expectations, compiler Rust products, production packages/provenance and language-1.2 verification records. Sampled array audit records match the stated axiom sets. This credits committed product presence and sampled bindings, not complete byte-correctness or final acceptance.

Review5875205367's SM39 coordination and allocation/exclusivity findings are unchanged, and the preserved pre-implementation R3 receipt remains unlinked. Please resolve those existing items without repeating the initial failure or completed producers. Current full CI43283951023 is active; earlier same-head43283909904 was canceled and is excluded, while exact-head repro43283909701 passed. The ready flag does not clear the remaining findings or final gate. Existing owner retains implementation and qualification.

@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.

Current hosted failure narrowed to a small downstream API correction. In full CI43283951023, rest job131413852835 checked out4233f519 (859b52 merged into26e5) and finished290 passed/1 failed. conformance_sm_16 fails compiling the embedded downstream snapshot consumer: Rust E0004, &SnapshotType::Array { .. } is not covered in its exhaustive type visitor.

In crates/conformance/src/support.rs, embedded fn ty around190, add explicit handling for the array element, analogous to List/Set, preserving the exhaustive match rather than adding a wildcard. This is a native downstream API compilation failure, not evidence of a Lean/kernel or array theorem failure. No cause is inferred for the other failed host jobs.

Existing owner retains the repair and resulting-head qualification. Preserve completed products/proof evidence; this diagnosis does not justify redoing their producers. Review5875205367's SM39 coordination, conditional allocation wording and linked preserved R3 receipt remain separate unresolved items. Exact repro passed; full CI has failures and cannot establish acceptance. One previously unread rest log was archived; no runtime, repeated log, rerun or patch was performed here.

…ay type

conformance_sm_16 compiles a crate outside the workspace that matches every public SnapshotType variant, so that a new variant is a compile error downstream until it is handled. It handled every type but the array.

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

@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.

Source repair credited atf943d9af: the sole changed line adds SnapshotType::Array { element } => ty(element) to the downstream consumer's exhaustive visitor, exactly addressing review5881767778's SM16 compilation omission without a wildcard. Existing generated products are preserved; no producer replay is needed for this source-only correction.

Current CI43333060260 and repro43333059785 own this head's qualification; the prior859b failed run is not acceptance. The unchanged SM39 allocation, conditional allocation wording and preserved R3 receipt items from review5875205367 remain separate. Source correction is established; hosted SM16 success/full final-head acceptance is still pending. Existing owner retains the work; no tests or logs were repeated.

afflom commented Nov 13, 2026 •

Copy link
Copy Markdown
Member

Current owner delivery handoff — 2026-11-15, exact14f21e8eda3a27ec6627de1d484bad47ec5a1bd8.

First concrete integration blocker: records run43703750928/job132780277777 checked out14f and stopped at10:27:33 with LLC0101: examples/arrays/lexlean.toml:10 still names leanprover/lean4:v4.32.1, while this head's root pin requires4.32.2. Align the arrays-owned project pin and review the affected fixture/product inputs. This is a configuration refusal before records, not evidence of an array theorem failure. One original log is retained (SHA256238e0da9f6659756b25d3b4992f1e099117f0ae97525a46a2e6d4abe5ac27d46); do not refetch/recreate it or automatically retry.

The existing smallest owner batch remains unresolved:

  • Apply the coordinated SM40 allocation across authoritative register, official projections, scenario, named test/dispatch and proposal. Current model/ids.toml:743, conformance.rs:614, semantic-ir.feature:278 and proposal still use SM39. Preserve historical evidence under its original identifier and the earlier reservations.
  • Qualify allocation claims in docs/proposals/verified-arrays.md:21,97,183: exclusive mutation versus copying and the Option-allocation assertion.
  • Link the already-preserved original R3 failure; citing the rule is insufficient. Do not recreate the initial failure.
  • Update body/proposal history for the current composition and later adopted output, while retaining prior Lean4.32.1 observations under their original binding.

Current14f CI43703753902 has1PASS/2FAIL/15SKIP; repro43703753872 failed; producer43703750928 has3FAIL/3SKIP and no delivered output. Only the records failure above was diagnosed. Old586178 full43600117408 attempt1 (70 jobs/772 steps) and exact repro43600117363 remain valid at their original head; they do not qualify14f.

Complete the corrective source batch in the existing owner lane, then disposition exact resulting output and final qualification with actual merged-tree equality. No competing patch/writer/gate or automatic retry was started. Bounded independent receipt0a2cb0cd5f632a24ff2e8d56f50946ce45282d0c18981c81a18bc67faff7009c; capped comparison is not a full1169-path audit.

CHANGELOG.md keeps both entries. Every generated record that conflicted is taken from the release line, and every lock and record names the combined language-1.2 identity; the regenerate workflow rewrites the records next.

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

@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.

Current-base integration credited: 8ae2de3 has ordered parents f943d9a + actual release b8b6e86. Prior array/source/proof evidence remains scoped to its recorded inputs; the old f943 full70/70 gate is not acceptance for this merged preparation.

The new regeneration blocker is concrete and small. In run43397143569, terminal oracles job131790722158 checked out this exact head. Golden writing completed, then fixtures-write stopped at tests/negative/array-under-1.1: check exited2 with LLC0102, “the compiler semantics changed since locking”, while case.toml expects1. The intended language-1.1 array refusal was never reached. Repair this fixture's lock through the official current language-1.1 lock path, preserving expected exit1/LLT4001; accepting stale-lock exit2 would weaken the fixture. One previously unread job log is retained; do not refetch it or reproduce the failure.

The existing owner should disposition the still-running production-export/records jobs and any retained outputs before choosing the necessary continuation. Successful write messages alone do not establish captured/reviewed artifacts. No blanket regeneration, cancellation or extra gate was initiated.

The selected authored-source review also confirms the prior source obligations remain: SM39/test name and unchanged proposal still need the coordinated SM40 mapping from6843396283; qualify allocation/in-place claims by runtime exclusivity; link the preserved initial R3 receipt without recreating it. Include these known small corrections in your existing batch before final qualification. Current CI43397145883 stopped at the generated-file honesty check (with downstream aggregate failures/skips), while repro43397145866 passed; no further CI cause is inferred. Existing implementation ownership is unchanged.

… identity RB-24 moved

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

@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.

The stale-fixture-lock finding from5888664727 is structurally resolved. This complete one-file, one-line delta changes array-under-1.1's compiler-semantics binding11732→e701, matching the current language1.1 identity, while preserving its language, source, configuration, pins and expected exit1. It does not accept LLC0102/exit2 or weaken the intended LLT4001 refusal.

The existing regeneration43406816533 now reports its oracles job successful; that completed stage is credited. Records and production-export remain active. Preserve and disposition those outputs before any necessary continuation; no old log, failure reproduction or blanket producer replay is needed for this repair.

SM40 allocation, conditional allocation/in-place wording and the preserved initial R3 receipt link remain unchanged in this lock-only delta. Keep those known corrections in the existing owner batch. Current PR CI43406816131 failed and exact repro43406816162 passed; full final-head qualification/product acceptance is still pending. No competing patch, runtime, writer or gate action was performed.

afflom commented Nov 13, 2026

Copy link
Copy Markdown
Member

The existing f20 regeneration43406816533 is now terminal SUCCESS,5/5 jobs. Its final patch artifact14033719483 (regenerated-f20bcc5096de6573f489fa5babafad0cfd9e016a) is present and nonexpired, SHA256463c6653bce168eeef3baec474cab2808a8172f3cca13e123f99a8444b9d80f2; combined/records/production/oracles artifacts are also retained. No new producer is needed merely because publication remains pending.

The next owner step is to disposition/adopt the exact-source output, complete the previously coordinated SM40 allocation and conditional allocation wording/R3 linkage, then qualify the resulting head. Currentf20 remains unpublished-product source and its PR gate failed; the successful regeneration does not substitute for final-head acceptance. The one-line fixture lock repair remains credited by5889729042; no old failure, log or completed writer was repeated.

Release/v0.4 has since gained workflow-only #516/15d66a; preserve actual history and determine resulting-source impact before any necessary further work.

…ty after merging RB-24

The patch of the regenerate workflow, run 43406816533 on regenerate/verified-arrays-3 at f20bcc5 (Linux x86-64, pinned Lean 4.32.1). No bootstrap count moved.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
… follow RB-24's row (495)

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

@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.

The complete five-file delta fixes the495-ID census in documentation and RP07/spec-link assertions. Model rows, array implementation, schemas, locks and adopted products are unchanged; their prior scoped credit is preserved. This retires the census mismatch structurally without claiming the failed c184 gate passed or repeating its checks.

The already-coordinated SM40 mapping, conditional allocation/exclusivity wording and link to preserved original R3 evidence remain unchanged. Please complete these focused source/evidence corrections together; do not recreate the initial failure. The body also still describes historical products instead of the c184 adoption.

Actual release is now b4ac72e, including #499 after #516; current head is conflicting. Reconcile that actual history and the check-only constant/census impact before resulting-head qualification, preserving applicable adopted products and avoiding a blind identity wave. The earlier516-only impact assessment does not cover the new #499 composition. Existing owner retains implementation; no producer, log, test or CI action was repeated.

…y build) into the arrays branch

Source conflicts: the runtime registry, backend/semantic.rs import selection, and both schemas now carry Arrays and Regions together (seven native runtime modules). The register is 497 IDs (493 base + RB-24 + RP-19 + PF-24 + SM-39); every count assertion, README line, and SPEC.md prose total follows. Every generated record that conflicted is taken from the release line, and every lock and record is rewritten to the combined language-1.2 identity; the regenerate workflow rewrites the records next.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

@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.

Actual-base integration is credited at155cb46f: ordered parents0eafe + fd7 preserve accepted release history, and the registry/backend/schema composition carries Arrays and Regions together. Current sampled records carry the combined bd29ae01 identity, but they remain provisional baseline records pending regeneration43475779393; prior c184 adoption remains evidence for its original inputs. Exact repro43475785020 passed. CI43475785012 failed generated-file freshness and the downstream host aggregates, with14 skips; no further cause or full acceptance is inferred.

The small owner corrections remain unchanged: apply/disposition the coordinated SM40 allocation from6843396283, make the proposal's allocation statements conditional on exclusivity, link the preserved original R3 failure, and update the stale body/proposal census and product history. Do not recreate the original failure or repeat producers to repair prose. Complete these in the existing batch, disposition the current combined outputs, then qualify the resulting final head. Independent receipt pr493-155-owner-delta credits integration without clearing those findings; no competing patch or repeated check was started.

Regenerates all generated artifacts (production export, verification
records, bootstrap oracles) after the second release/v0.4 merge that
reconciled arrays (SM-39) with decision regions (PF-24) and the
check-only build (RP-19), bringing the register to 497 rows.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

@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.

Bounded publication readback: existing regeneration43475779393 succeeded for155cb46, and60cdeb03 is its direct output-adoption child. All inspected top-level authored .github/crates/model/language/schemas/features/xtask/SPEC/VERIFICATION objects are identical to155cb46. That preserves prior source credit and records adoption; it is not a complete product audit or final-head acceptance (commit-file listing caps at300).

The small source/evidence dispositions in5896121728 remain unchanged: coordinated SM40 allocation, conditional allocation/exclusivity prose, original R3 evidence link and current body/census/product history. Complete them in the existing owner batch without recreating the initial failure or repeating a successful producer just for publication/prose. Review actual be7/6a02 release changes for resulting-tree applicability, then qualify the final source/product head. No competing implementation, new identity wave or gate was started.

…unt rewrites) into the arrays branch

Only CHANGELOG.md conflicted: the arrays entry stays first, #517's entry
follows. No language data, schema, or fixture changed upstream, so the
language-1.2 identity and the regenerated files of 60cdeb0 stand; the
register is still 497.

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

@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.

The actual-base reconciliation is now credited:7a3e949c has ordered60cdeb03 + actual6a02 parents. Its complete16-path incoming delta is automation/count-tool history; product, language, schema, model and proposal trees remain identical to the adopted result. Keep those valid outputs—this merge indicates no new identity writer.

The already coordinated SM40 allocation, conditional allocation/exclusivity wording, preserved original R3 receipt link and current body/proposal history remain unchanged. Please finish those focused source/evidence dispositions together; do not recreate R3 or regenerate merely for prose/publication. Current CI43516259218 is queued and exact repro43516259269 passed. Prior green gates remain scoped to their original trees; final resulting-head acceptance is still owed. No tests, logs, gates or producers were repeated.

humuhumu33 and others added 2 commits November 14, 2026 04:10
…s branch

Only CHANGELOG.md conflicted: the arrays entry stays first, #498's follows.
#498 changes no language data, schema, or lock, so the language-1.2 identity
and the regenerated files stand; the register is still 497.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…the arrays branch

CHANGELOG keeps both Unreleased entries: the arrays entry and #520's.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
afflom pushed a commit that referenced this pull request Nov 15, 2026
SM-39..SM-41 are claimed on release/v0.4 (#493), so decimal formatting
isolation is registered as SM-43. The generated projections are release/v0.5's,
rewritten by the landing regeneration.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
humuhumu33 and others added 2 commits November 15, 2026 10:22
…#535, #536) into the arrays branch

The generated outputs that both sides rewrote take release/v0.4's copy, or
are removed where the line no longer has them; regenerate.yml rewrites the
whole set from this tree.

Lean moves to 4.32.2 with #518, so the array authority follows the other Lean
rows: LEAN-ARRAY-4-32-1 becomes LEAN-ARRAY-4-32-2, cited at f3b06c705e6c.
The three files its statement rests on (src/include/lean/lean.h,
src/Init/Prelude.lean, src/Init/Data/Array/Set.lean) are byte-identical at
f054605aea4b and f3b06c705e6c, so the statement carries over unchanged.

runtime-1.2/library.toml keeps the line's toolchain and the branch's Arrays
module. LANGUAGE_1_2_SEMANTICS_ID is the merged tree's, as RP-10 recomputes it.
CHANGELOG and VERIFICATION keep both sides' entries.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The merge moved the language-1.2 compiler-semantics ID from the line's
4b3ed237 to 946c112c, the ID of the tree with the arrays runtime module; it is
carried to every lock, provenance record and manifest that named it (never
*.md), as #518 did for 4.32.2. examples/arrays and the five array negative
fixtures pinned leanprover/lean4:v4.32.1 and were refused with LLC0101 before
reaching their case; they now pin v4.32.2 and are relocked with the built
compiler. Every lock the line keeps current checks (324 of 337; the other 13
are negative fixtures whose locks are refused by design), and each array
negative fixture reports its expected code.

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

@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.

Bounded follow-up at 64331b0, preserving the earlier 14f receipt and failure log without replay.

Credited: arrays and all five negative fixture project/toolchain pins now match Lean 4.32.2. The complete parent-to-head inventory covers 630 changed blobs (266 locks, 330 provenance records, 21 expected manifests, six configs, six Lean pins and one README), obtained from seven non-truncated subtree comparisons. Code, programs, runtime, schemas, register, scenarios, proposal, SPEC and VERIFICATION are unchanged by blob identity.

Output acceptance remains pending: examples/arrays/expected/build/manifest.json still records Lean 4.32.1, old compiler semantics bd29ae… and old configuration digest 8cf22d…; the repaired lock records 946c112c… and configuration 05159805…. Arrays build/verify products were not changed. Existing producer 43745353103 is still active at this exact head; its owner should disposition its checkout/results and adopt applicable completed outputs. Do not start duplicate regeneration.

The existing source owner still owns the coordinated SM40 register/scenario/test/dispatch batch (model/ids.toml:743, conformance.rs:614, semantic-ir.feature:278), conditional allocation prose (verified-arrays.md:21,97,183), the original preimplementation R3 evidence link, and PR title/body updates. Preserve the coordinated SM39/SM41 allocations. The body still describes SM39, Lean 4.32.1 and historical run 43241009907; distinguish that original evidence from current integration and producer history. Never recreate a red receipt.

Receipt: pr493-6433-delta-review-20261115.json, SHA256 cc8cbc9a4f0571b20a467f035515caf45291c0f49dc684bddf044f6a8afe0867. This credits a source repair, not runtime acceptance. Final composed-head qualification and actual merged-tree equality remain required. No tests, writer, retry, cleanup or log refetch performed.

The write recipes of SPEC.md §9.2, run by regenerate.yml (https://github.com/UOR-Foundation/lexlean/actions/runs/43745353103),
rewrote 5648 paths.

The counts writer reported:

    counts: no count moved

@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.

The existing producer43745353103 completed all six jobs, including combine/counts/deliver. Current0df6 is its direct output-adoption child of6433. Complete new-only path/hash classification finds only generated products/provenance; authored source and workflow subtrees are unchanged. Arrays manifest now records Lean4.32.2 and the current946c112c… compiler /05159805… config identities, superseding the stale-manifest finding. No duplicate producer is needed to adopt outputs already present.

The next blocker is concrete and smaller: exact-head CI43753524838 honesty job132932714429 passes freshness (25 oracles,278 recorded inputs,164 exports) and validate-model (498 IDs), then fails RP-07: “the §31 table has498 rows, not497”. xtask/src/spec_links.rs:64 still hardcodes497. Existing source owner should reconcile that expectation with the registered/spec rows while completing the outstanding coordinatedSM40/prose/original-R3-link/body batch. Preserve register/spec bijection and existing ID ownership; do not weaken the check or recreate red history. Then qualify the final resulting head.

Receipt pr493-0df6-delivery-review-20261115.json SHA256215dfc0b7b30383ce7660312e8527859f7c9be3d07e5c46a3da09745cfc0c89c. One new failure log retained, SHA256fef7a1d6c28621542ade2a84ee20a73cca62b9b0404f83ca7f3d622767976cee. Artifact14177306565 metadata is linked to6433; ZIP bytes were not compared, so no ZIP equality or full acceptance claim. Prior producer/pin evidence reused; no tests/writer/retry/cleanup performed.

The line reached 497 with CL-25 while the arrays branch reached 497 with
SM-39 on the same base; the merge holds both, and RP-07's row count, RP-06's
table length, SPEC.md's total, the README and the changelog say 498.

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

@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.

The complete five-file0317 delta addresses the concrete row-count blocker: RP-06/RP-07 and SPEC/README/CHANGELOG consistently move497→498 while retaining the register/spec bijection checks. This matches the already observed498 rows/IDs at0df6; the delivered products are preserved. No producer repeat is justified by this count-only repair.

The coordinated SM40/prose/original-R3-link/current-body batch is unchanged and remains with the existing source owner. Live API currently returns no check runs or workflow runs for0317; the failed0df6 gate does not qualify this new head. Finish the source batch, then supply final-head qualification and actual merged-tree equality.

Bounded receipt pr493-0317-pr553-893c-repairs-20261115/receipt.json SHA256fceee3df7b95946a1473a080ed42094c98c14c5aff2cea277b02902d6dbdfd2e. Prior failure/product evidence reused; no tests, writer, retry or log refetch.

The arrays entry stays under Unreleased above the cut's 0.4.3 section; the
generated manifests keep this branch's copies until regeneration writes them
at 0.4.3.

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

@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.

Bounded e42 integration readback: exact merge parents0317 +external releasefa271; complete13-path delta is release/version integration and associated manifest/hash changes. Arrays implementation/register/scenario/proposal source is unchanged, so the existing SM40/prose/original-R3-link/body obligations remain.

Product applicability is now version-specific: examples/arrays/expected/build/manifest.json still reports compiler0.4.0 while the workspace is0.4.3; its semantics946c… and Lean4.32.2 remain unchanged. The merge message explicitly keeps branch outputs pending regeneration. Preserve completed6433/0df producer evidence at its original binding; it does not establish freshness of the new versioned manifest. Existing producer43768641714 is active, with CI queued/repro passed. Its owner should disposition the resulting outputs and reconcile any later source batch before final-head qualification. No additional producer or retry is requested.

Receipt pr493-e42c-release-merge-review-20261115.json SHA256738b685e254cdd4aadb6de80729b144ff497a4979a2b66040c37b0c12c584a24. Large uor-atlas manifest patch was unavailable; no exhaustive product-content claim. Release ancestry does not lift #431/oldTC03 STOP. No execution, writer, log refetch, cleanup or CI control.

The write recipes of SPEC.md §9.2, run by regenerate.yml (https://github.com/UOR-Foundation/lexlean/actions/runs/43768641714),
rewrote 22 paths.

The counts writer reported:

    counts: no count moved

@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.

Delivered-output adoption credited atd0b9ff78: exact bot child ofe42c447b, linked producer43768641714 completed all six jobs including counts/delivery. Final artifact14188029454 is nonexpired and bound toe42, digest08b9cf07fccc06b6577e56aedadd9855077646cf48cde2d23c20c707cd9c3038 (metadata custody, not ZIP-byte equality).

Complete22-file delta consists only of expected/build/manifest.json files. All21 available patches change compiler.version0.4.0→0.4.3 with semantics/other fields unchanged; the large bootstrap patch is unavailable and classified by path only. Arrays' manifest now reports0.4.3, resolving the prior workspace/version mismatch. Authored implementation is unchanged; no producer or gate was repeated.

Remaining existing-owner work is the previously scoped SM40/prose/original-R3 linkage/current-body batch and final-head qualification. CI43783410629 is active at inspection, not final acceptance; actual merged-tree equality remains mandatory. Receipt /workspace/work/pr493-d0b9-product-adoption-20261116.json; SHA25689aa63392eb2e6e83e4d5e57f51755593aba594666f51f82fdd263d6c4118b4f. No issue closure or stopped-work adoption.

@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.

Completed CI 43783410629, attempt 1, is credited at d0b9ff7: 66 successful jobs and 791 successful steps, including same-run honesty and Atlas join. Three Intel jobs were skipped. The new plan log binds checkout d0b9 to tree 61c260a7. This is successful execution of the planned jobs, not full platform qualification.

The exact-head Actions listing contains only that CI run; no d0b9 reproducibility run was observed. Earlier e42 repro evidence retains its original input binding. Existing owner: identify applicable Intel/repro evidence, complete the retained SM-40 allocation/projection/test batch, allocation prose and original R3 evidence references, and reconcile the body’s SM-39/Lean 4.32.1/old-run wording. Then qualify the resulting final composition and actual merged tree. No blanket rerun is requested.

The completed producer and 22-manifest version repair remain credited. Receipt: /workspace/work/pr493-d0b9-terminal-gate-20261116.json; SHA256 f7d7ec5382cd15fb520ffd32e869a91a9a6f56ff5f47ca82a1b3378701751a7a. All standing holds remain.

humuhumu33 and others added 2 commits November 16, 2026 09:09
…arrays branch

PF-26 and the arrays both move the language-1.2 identity from the 0.4.3 base;
the combined tree's ID, b78cb38d…, replaces each side's in the constant, the
locks and the records, and the register is 499 IDs: PF-26 joined the line as
its 498th, SM-39 is the 499th. A conflicted generated projection takes the
line's side, and one only this branch named is dropped, for regeneration to
write.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The write recipes of SPEC.md §9.2, run by regenerate.yml (https://github.com/UOR-Foundation/lexlean/actions/runs/43837359847),
rewrote 5647 paths.

The counts writer reported:

    counts: no count moved

@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.

Bounded follow-up at delivered e4b90d4 (source f3cb74b). The merge has ordered parents d0b9ff7 + actual release/v0.4 0541025; the output commit directly parents that merged input. Producer 43837359847 attempt 1 completed all six jobs successfully, including capture/combine/counts/delivery. Its final artifact 14216279959 names exact f3cb74b (digest sha256:7be18f35585a77849c17f6c0eac507028458b1303044a7aaa9a9bc6141f4061f). This credits delivered product custody, without archive-byte equality or final acceptance.

Both compare inventories hit the 300-file cap. A bounded hierarchical tree comparison instead establishes that f3cb→e4b changes only 28 expected-output subtrees, 19 compiler negative JSON fixtures, and 330 package/provenance pairs; 1,229 identical boundary hashes preserve source/config/model/scenario/docs/runtime trees. No ZIP or broad product-content audit was performed.

The remaining owner batch is retained: authoritative arrays row/scenario/named test still use SM-39, not the coordinated SM-40 allocation; the proposal still says 495 rows and assigns #434 the next free ID, while the integrated register/projection expects 499. The unchanged proposal still says a write allocates nothing in D1 despite its conditional exclusive/copying explanation; the original preimplementation R3 failure reference remains outstanding. The current title/body still cite SM-39, conformance_sm_39, Lean 4.32.1/LEAN-ARRAY-4-32-1 and producer 43241009907 rather than the current composition and preserved delivery history.

Manual final-head CI 43851090792 remains incomplete; observed check metadata includes three Intel skips. Prior d0b9 CI and older repro retain their original scope. Existing owner keeps source repair, applicable Intel/exact-head repro/full gate, and eventual actual merged-tree obligations. No producer/gate/retry/build/test repeated and no merge or source-only acceptance. All #316 frozen905/BOTH/initialR3/errors/sevenSTALE/failed-success protections, #431/oldTC03 and R25 STOP, future #338/#438/#541 holds remain unchanged.

@humuhumu33
humuhumu33 merged commit 2e1ff2f into release/v0.4 Nov 16, 2026
147 of 148 checks passed

@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.

External landing readback: #493 merged at 12:57:43Z as 2e1ff2f, with ordered parents 0541025 + e4b90d4. Actual tree b5a55394 equals the reviewed delivered-head tree, so no fresh source changes resolve the retained findings. Prior producer/product custody remains credited.

The complete latest-attempt inventory for manual CI43851090792 attempt2 is now available: 74 unique jobs, 71 successful jobs, three Intel skips, and 831 successful steps, including honesty and Atlas join. The new plan checkout log binds exact e4b90d4; its inherited 10:41 execution timestamps precede attempt2 and do not show all jobs reran. Original attempt1 remains separate (70 success, one structural-grammar failure, three Intel skips). This is planned-job technical credit, not all-platform qualification.

Actual-merge reproducibility43871322184 reports success at exact2e1f; current actual CI43871322239 remains incomplete. Bot-head repro43851114709 failure is separate, not overwritten by this later successful actual-tree result. No failure cause inferred.

Full acceptance remains false. Existing owner retains coordinated SM40 register/projection/scenario/test/dispatch/proposal allocation preserving SM39/SM41 history; proposal495 versus499 census; conditional exclusivity/allocation prose; preserved original preimplementation R3 reference; and current title/body/product history. External landing does not retire these obligations, authorize issue closure, or waive the applicable Intel/final actual qualification and forward-history requirements. No producer, build, test, retry, cleanup or GitHub mutation performed. All #316 protections, #431/oldTC03 and R25 STOP, future#338/#438/#541 and #468 holds remain unchanged.

@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.

The actual merged-tree technical gate is now complete. CI43871322239 attempt1 at actual2e1ff2f5 passed all77 unique jobs and878 steps, with zero skips or failures. This includes same-run honesty, Intel construction and all four lanes, and Atlas join/native-verification assertion. The new plan checkout log binds exact2e1f; run metadata and retained landing evidence bind treeb5a55394, equal to the reviewed delivered head. Exact-actual reproducibility43871322184 also passed.

This resolves the previously incomplete actual-tree technical gate. Earlier manual43851090792 attempt2 remains separately scoped to71 successful jobs/831 steps plus3 Intel skips and inherited timestamps; its original failure and bot-head repro failure remain preserved.

Full acceptance still requires the existing owner's coordinated SM40 register/projection/scenario/test/proposal allocation, census495 versus499 and allocation-prose corrections, original preimplementation R3 reference, and current PR-body/history disposition. Forward index.lock remains a separate blocker. No issue closure, retrospective premerge approval, retry or STOP waiver is implied.

Receipt pr493-2e1f-terminal-gate-20261116.json; SHA2565b6e4c566e3dc10555fdf0c048dbad9178c7ad60344ef6283bfd69733286a2e6.

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