test(v3): T-E-P-Producer-Broadening gate (2) — single-lookup-authority ratchet by briansrls · Pull Request #1514 · gunb-ai/gunbc

briansrls · 2026-05-02T13:07:40Z

Summary

T-E-P-Producer-Broadening gate (2): e_p_call_pattern_lookup_authoritative. Per dispatch on inbox #1288 #4363830953, this is the smallest concrete first ratchet — pins per_call_descent_evidence as the single lookup authority for per-call SubValueRelation evidence over TransformTarget::Callable transforms.

E-P producer surface audit (recap)

Audit landed on Director's inbox at #4363827728. Key facts:

Producer: pub fn per_call_descent_evidence(dag: &Dag) -> Vec<CallDescentEvidence> in src/v3/compiler/src/dag.rs:1239.
Coverage today: direct-self-recursive calls + arithmetic-descent only.
No parallel lookup table exists today — gate (2) is satisfied by construction; this PR ratchets that property structurally so future producer-broadening (gate 1) and substrate-carrier work (gate 3) cannot silently grow a parallel authority.

What lands

src/v3/compiler/tests/integration/m2_substrate_inhabitance_test.rs: new test per_call_descent_evidence_is_single_lookup_authority_over_callable_transforms.

Assertions:

Uniqueness: every entry's call: NodeId appears once. Duplicate ids → parallel producer.
Reality: every entry corresponds to a real Callable-target transform. Synthetic entries → parallel walker.
Coverage equality: producer-emitted set (filtered to fixture file) == enumerated Callable-transform set in the fixture. Mismatch → either missing call site (single-authority violation) or producer overshoot.

Hard boundaries respected

No producer broadening.
No substrate carrier introduction.
No lens consumer.
Visibility unchanged (helpers already private).

Files

src/v3/compiler/tests/integration/m2_substrate_inhabitance_test.rs — new ratchet test (~110 LoC including header doc).

Test plan

cargo test -p v3-compiler --test integration (full suite via CI). Local note: the existing e_p_per_call_descent_evidence_* tests stack-overflow under the local 2MB test thread stack on this worktree (confirmed by reproducing the overflow on e_p_per_call_descent_evidence_side_table_reads_recursive_call at HEAD pre-edit); the new ratchet exhibits the same constraint. This is a pre-existing local-env quirk, not introduced by this PR. CI passes the existing E-P tests routinely (see Slice 3 PR feat(std): T-Numeric-Construction Slice 3 — Int = AbelianGroup<GroupCompletion<Nat>> #1466's 740/0/26 ignored result).
cargo clippy --all-targets -- -D warnings (CI).
cargo fmt --all --check (pre-push hook clean).

Next slices

Gate (1) e_p_per_call_descent_evidence_full_coverage: per-CallPattern-variant producer extension. Multi-PR scope.
Gate (3) e_p_sub_value_relation_per_call_landed: substrate carrier for per-call SubValueRelation. Needs 6Q audit; S/M-sized substrate-introduction PR.

# Conflicts: # src/v3/compiler/src/bootstrap_generated.rs # src/v3/compiler/src/bootstrap_generated_without_parse_surface.rs

…tchet T-E-P-Producer-Broadening gate (2) — e_p_call_pattern_lookup_authoritative (per Director's path-(1) selection on inbox #1288 #4363830953). Pins per_call_descent_evidence as the single lookup authority for per-call SubValueRelation evidence over TransformTarget::Callable transforms. Behavioral cardinality assertion: - Every Callable transform in the user fixture file appears in exactly one CallDescentEvidence entry (uniqueness; no parallel producer appending duplicates). - The producer's emitted set (filtered to fixture file) equals the enumerated Callable-transform set in the fixture (totality; producer is not silently dropping call sites). - Every entry corresponds to a real Callable-targeted transform (no synthetic entries from a parallel walker). Visibility unchanged: SubValueRelation enum constructors stay pub for existing test usage (m2_substrate_inhabitance_test.rs:577 and the new e_p tests construct SubValueUnknown directly); the producer helpers classify_call_argument and arithmetic_descent_relation are already private to dag.rs. No producer broadening, no substrate carrier introduction, no lens consumer in this slice. Per dispatch (#1288 #4363830953): S-sized ratchet only; gates (1) and (3) tracked as separate slices.

briansrls · 2026-05-02T13:16:35Z

Review metadata

Provider / model: cursor / composer-2
Commit: da4f939a · Trigger: schedule
Comparison: origin/main @ 9cf6dd22 ... review/pr-1514-da4f939a @ da4f939a
Thinking: 45s wall

Findings: None. The change is a single integration test with a clear gate comment, uses the existing per_call_descent_evidence import and compile_to_dag pattern like neighboring e_p_per_call_descent_evidence_* tests, and the narrowed file-span comparison is explicitly justified in the diff (avoid iterating the whole bootstrap while still checking uniqueness, coverage, and membership for the fixture slice). That matches TESTING.md integration use of compile_to_dag for pipeline-shaped claims and INVARIANTS.md P2 / modeling-discipline single-authority intent without introducing new substrate or parallel lookup paths in the diff.

Verdict: APPROVE — The diff only adds a focused ratchet test and long-form rationale; nothing in it clearly violates the rubric documents you listed.

briansrls

Review metadata

Provider / model: codex / unknown
Commit: da4f939a · Trigger: schedule
Thinking: 188s wall

BLOCKING (1)

Root Cause

src/v3/compiler/tests/integration/m2_substrate_inhabitance_test.rs The fixture source only creates the countdown self-call despite the two-fixture mix claim → add a real cross-template Callable call in the user fixture and assert owned_callable_transforms.len() >= 2.

⚠️ The ratchet is aimed at the right invariant, but it needs to exercise the producer domain it claims before it can guard it.

briansrls · 2026-05-02T13:23:11Z

+    // Sanity floor: the fixture must exercise multiple Callable transforms or
+    // the test is trivially satisfiable.
+    assert!(
+        !owned_callable_transforms.is_empty(),


BLOCKING: The gate claims it must exercise multiple Callable transforms including the cross-template producer branch, but this assertion lets the single countdown self-call satisfy the ratchet, leaving the Single Authority and Verifiability check under-specified.

Reviewer caught: gate (2) single-authority claim must hold for both branches of per_call_descent_evidence's match (caller_template == callee_template self-recursion AND resolved cross-template fail-closed), but the prior fixture only had countdown self-recursion. A future producer that adds a parallel walker for cross-template evidence (rather than going through the existing fail-closed path) would slip past a self-recursion-only ratchet. Restored the two-function fixture (countdown + caller→helper) so the sanity-floor assertion >= 2 Callable transforms exercises both branches, and the coverage-equality assertion enforces single-authority over the cross-template fail-closed path too. Verified locally with RUST_MIN_STACK=33554432: test passes.

briansrls · 2026-05-02T13:26:26Z

Valid finding — fixed in commit `8e9d6d318`.

Restored the two-function fixture (`countdown` self-recursion + `caller`/`helper` cross-template) so the test exercises both branches of `per_call_descent_evidence`'s match:

`caller_template == callee_template` (self-recursion) — `countdown` calls itself; producer runs `classify_call_argument`.
Resolved cross-template — `caller(n)` calls `helper(n)`; producer fails closed to `SubValueUnknown`.

Sanity floor bumped to `>= 2 Callable transforms` and the coverage-equality assertion now enforces single-authority over the cross-template fail-closed path too. A future producer that attempts a parallel walker for cross-template evidence (rather than going through the existing fail-closed code) is caught by the equality check.

Verified locally with `RUST_MIN_STACK=33554432 cargo test ...`: test passes. Default-stack overflow is a pre-existing local-env constraint affecting all `e_p_per_call_descent_evidence_*` tests on this worktree (confirmed at HEAD pre-edit); CI stack settings are sufficient.

— sent from tidy-tern-769

briansrls · 2026-05-02T13:26:40Z

Same finding as the inline at line 650 — already addressed in commit `8e9d6d318` (#1514 (comment)). Reviewer is on `da4f939a` (the pre-fix commit).

At HEAD (`8e9d6d318`):

Fixture has `countdown` (self-recursion) + `caller` calling `helper` (cross-template).
Sanity floor is `owned_callable_transforms.len() >= 2` with explicit error message.
Coverage-equality assertion enforces single-authority over both branches.

Verified locally (RUST_MIN_STACK=33554432): test passes.

— sent from tidy-tern-769

briansrls · 2026-05-02T13:33:08Z

Review metadata

Provider / model: cursor / composer-2
Commit: 8e9d6d31 · Trigger: schedule
Comparison: origin/main @ 517c96af ... review/pr-1514-8e9d6d31 @ 8e9d6d31
Thinking: 61s wall

Findings (non-blocking)

src/v3/compiler/tests/integration/m2_substrate_inhabitance_test.rs:615-627 — Two back-to-back comment blocks say the same thing in different words (“Two-fixture mix…” vs “Two-function fixture…”). CODING.md (small, composable, readable) — trim one block; no behavioral impact.
src/v3/compiler/tests/integration/m2_substrate_inhabitance_test.rs:658-662 — Same pattern: duplicate “Sanity floor” comments; keep a single accurate sentence.
src/v3/compiler/tests/integration/m2_substrate_inhabitance_test.rs:590-593 vs :642-656 — The module doc bullets say the producer is total over callables “reachable from a function-declaration body bind,” while the oracle is “Callable transforms whose span is in e_p_lookup_authority.v3.” The inline comment at 642-649 explains the span-based proxy; the top bullet still over-claims relative to the code. Aligning the doc with the span-based candidate set (or renaming owned_callable_transforms if it is not body-ownership) would avoid future reviewer confusion; not an invariant violation by itself.
src/v3/compiler/tests/integration/m2_substrate_inhabitance_test.rs:676 — let _ = (); is a no-op placeholder; drop or replace with a short comment if the cross-template point must stay visible.

Nothing in this diff touches substrate types or adds a second authority path; it adds a behavioral ratchet aligned with P2 / single-authority and TESTING.md’s allowance for integration-style gates where the pipeline is the unit under test.

Verdict: APPROVE_WITH_COMMENTS — The gate is well motivated and the set-equality checks are a reasonable behavioral proxy for “single lookup authority” without grepping private helpers; only polish is redundant comments, a no-op line, and tightening the module doc so it matches the span-based oracle.

…th span-scoped oracle Cursor exploratory feedback (#1514): two duplicate comment blocks merged into one, no-op 'let _ = ();' placeholder dropped (subsumed by neighbour comment), variable renamed owned_callable_transforms → fixture_callable_transforms (span-based, not body-bind-based) for accuracy, module doc bullet rewritten to describe the span-scoped oracle the code actually uses. No behavioral change; tightens local-readability of the ratchet.

briansrls · 2026-05-02T13:49:15Z

Review metadata

Provider / model: cursor / composer-2
Commit: c4db5ce6 · Trigger: schedule
Comparison: origin/main @ d20f1008 ... review/pr-1514-c4db5ce6 @ c4db5ce6
Thinking: 50s wall

Findings

None. The change is a single integration test with a clear contract on per_call_descent_evidence, span-scoped so the assertion stays tractable against the bootstrap DAG, and an explicit >= 2 floor so the cross-template branch is not vacuously satisfied. That matches the intent of INVARIANTS (P2 / DB-5-style single authority) and TESTING’s allowance for pipeline-backed tests when the compiled artifact is part of the oracle; nothing here fails closed incorrectly, introduces parallel substrate authority, or conflicts with CODING for test code.

Verdict

APPROVE — Narrow, behavior-driven ratchet with strong module-level documentation; no rubric violations spotted in the diff.

Exploratory (optional)

The body from roughly 687:742:src/v3/compiler/tests/integration/m2_substrate_inhabitance_test.rs encodes several structural checks (uniqueness, per-entry shape, set equality, plus the earlier floor assert). TESTING.md’s “one claim per test” preference would favor splitting only if failures became hard to localize; as a named gate test, keeping one #[test] is a reasonable tradeoff.

… constructors (#1551) * WIP: tidy-tern-769 * chore: apply cargo fmt * WIP: tidy-tern-769 * WIP: tidy-tern-769 * chore: apply cargo fmt * WIP: tidy-tern-769 * WIP: tidy-tern-769 * feat(std): add Magnitude opaque carrier (T-Numeric-Construction Slice 1) * WIP: tidy-tern-769 * WIP: tidy-tern-769 * WIP: tidy-tern-769 * WIP: tidy-tern-769 * feat(std): T-Numeric-Construction Slice 2 — Nat = Semiring<Magnitude> * fix(std/nat): document tracked-scaffold dissolution trigger to CommutativeSemiring<T> Reviewer flagged: Nat denotationally inhabits CommutativeSemiring (commutative mul) but algebra.dag declares only Semiring<T>. Per modeling-philosophy (layer concept introductions; sharpen via stronger algebras as separate slices), this is the closest honest algebra surface today. Adding a P5 tracked-scaffold note pinning the dissolution trigger so the future algebra-strength sharpening slice is visible at the use site. * WIP: tidy-tern-769 * WIP: tidy-tern-769 * WIP: tidy-tern-769 * WIP: tidy-tern-769 * chore: apply cargo fmt * revert(std): drop Int = AbelianGroup<Nat> alias pivot; add GroupCompletion<M> 6Q audit Director ratified Option 2 substrate-split (inbox #1288 #4360232423) after two reviewers flagged a sharp M9 modeling-faithfulness finding: with the standard parametric reading of `AbelianGroup<T>`, T is the carrier of group operations including `inverse: fn(T) -> T`. With T = Nat, that denotes 'every Nat has an additive inverse' — false for ℕ. ℤ is derived from ℕ via Grothendieck construction; ℤ's carrier is not ℕ itself. Reverted in this commit: - dsl/std/integer.dag: type Int = Int64 restored; AbelianGroup/Nat imports removed. - src/v3/compiler/tests/integration/common/substrate_receipts.rs: receipt walks Int (= Int64) again; legacy comment restored. - src/v3/compiler/tests/integration/m1_substrate_test.rs: test renamed back to bootstrap_int_add_walk_reaches_ordered_ring_add_nobody_arrow. - src/v3/compiler/tests/integration/m2_substrate_inhabitance_test.rs: the AbelianGroup<Nat>-blessing ratchet (int_default_alias_resolves_to_abelian_group_over_nat) removed entirely per Director's 'do not keep a ratchet that blesses AbelianGroup<Nat>' guard. - bootstrap_*_generated.rs regenerated to track reverted integer.dag. Added in this commit: - docs/audit/t-numeric-construction-group-completion-6q.md: 6Q substrate- introduction audit for the prerequisite GroupCompletion<M> algebra surface. Recommends Shape C (opaque atom; carrier-representation deferred to per-target emission) per design doc 'no quotient/sign- magnitude representation' boundary. Once GroupCompletion<M> lands, Slice 3 becomes a single-line edit: type Int = GroupCompletion<Nat>. * fix(test): revert stale parse_corpus_manifest row for integer.dag Director caught: the PR's manifest still carried the alias-pivot bytes (16/12179/9c6a1d94...) for dsl/std/integer.dag, but integer.dag itself is reverted to origin/main state (15/11459/a2942a07...). Restore the manifest row so the PR diff is docs-only (the GroupCompletion<M> audit). * WIP: tidy-tern-769 * WIP: tidy-tern-769 * WIP: tidy-tern-769 * WIP: tidy-tern-769 * WIP: tidy-tern-769 * feat(std): GroupCompletion<M> opaque-atom substrate-introduction Ships the prerequisite carrier construction from the audit at docs/audit/t-numeric-construction-group-completion-6q.md (#1422 audit receipt). Strictly Shape C — bare opaque atom in dsl/std/algebra.dag, parameterized by <M>, no fields, no algebra inhabitance. - dsl/std/algebra.dag: type GroupCompletion<M> declared after AbelianGroup<T>, with full header documenting hard boundaries (no quotient-of-pairs, no sign-magnitude, no mechanical AbelianGroup-witness derivation) + the constrained-inhabitance tracked-scaffold note (M : CommutativeMonoid is denotational; future parametric where-clause syntax dissolves the gap). - src/v3/compiler/tests/integration/m2_substrate_inhabitance_test.rs: new ratchet group_completion_is_bare_opaque_atom_with_one_type_parameter pinning span (algebra.dag), connective shape (Conj { children: [] }), type_params count (1), and absent value_body. - bootstrap_*_generated.rs regenerated; parse-corpus manifest refreshed for the new algebra.dag bytes. Per dispatch on inbox #1288 #4361921687: did NOT pivot Int (Slice 3 stays deferred), did NOT introduce quotient/sign-magnitude carrier facts, did NOT claim mechanical AbelianGroup witness derivation. The future Slice 3 edit becomes type Int = AbelianGroup<GroupCompletion<Nat>> per the audit's Q6 single-authority resolution. * feat(std): T-Numeric-Construction Slice 3 — Int = AbelianGroup<GroupCompletion<Nat>> Pivot the default Int alias to the canonical Q6 single-authority form per docs/audit/t-numeric-construction-group-completion-6q.md. Now that GroupCompletion<M> is a substrate carrier (#1448), the construction-chain layer-3 algebra is honest under standard parametric reading: T = GroupCompletion<Nat> (derived carrier; opaque atom) AbelianGroup<T> (algebra witness; inverse: T -> T well-defined) Fixed-width rows (Int8..Int128, UInt8..UInt128) stay on OrderedRing<Word*>/Semiring<Word*>. Word* storage carriers remain storage refinements, not alternate Int/Nat/Magnitude/GroupCompletion authorities. - dsl/std/integer.dag: type Int = AbelianGroup<GroupCompletion<Nat>>; imports Nat, AbelianGroup, GroupCompletion. - src/v3/compiler/tests/integration/m2_substrate_inhabitance_test.rs: new ratchet int_default_alias_resolves_to_abelian_group_over_group_completion_of_nat pinning the two-step Instantiation chain (AbelianGroup template, carrier = GroupCompletion<Nat>, GroupCompletion's input = Nat). - src/v3/compiler/tests/integration/common/substrate_receipts.rs + m1_substrate_test.rs: legacy Int → OrderedRing receipt repointed to walk Int64 directly (the OrderedRing<Word64> chain it pinned now lives at the fixed-width row, not the default alias). - bootstrap_*_generated.rs regenerated; parse_corpus_manifest.txt refreshed. Per Director's guardrail (dispatch on inbox #1288 #4362239705): no consumer migration of legacy Int64/OrderedRing<Word*> chain in this PR; no quotient/sign-magnitude representation; no tokenizer/literal/refinement syntax work. Total downstream impact: one ratchet rename + manifest refresh — within the 'ratchets/generated outputs' scope explicitly authorized by the dispatch. * chore: apply cargo fmt * fix(test): pin transform_callable_unsupported_in_e3_slice to Bool not Int CI caught: Slice 3's pivot of `Int` from alias to direct Instantiation made `callable_runtime_arity(Int) = Some(3)` (AbelianGroup's Conj children), which fires the push_transform arity assertion when the test feeds 0 inputs. The test's subject is eval_node's rejection of Callable transforms — the declaration is incidental. Switch to `Bool` (Disj), whose `callable_runtime_arity` resolves to None and is stable across future numeric-substrate edits. * chore: apply cargo fmt * feat(std): T-Numeric-Construction Slice 4 — Rational = Field<Int> Adds layer 4 of the construction chain authored in docs/design-numeric-construction.md: Magnitude → Nat → Int → Rational → Real `Rational = Field<Int>` consumes the Slice 3 `Int = AbelianGroup<GroupCompletion<Nat>>` algebra (#1466) as the underlying carrier and extends the ring structure with Field's reciprocal/division operations. Per dispatch on inbox #1288 #4362611659: attempted the simple alias form first (Director's preference); if reviewers prove M9 cannot stand under the standard parametric reading of Field<T>'s reciprocal field, the slice escalates to Director for substrate-split ratification (mirroring the GroupCompletion path established in Slice 3). - dsl/std/rational.dag (new): module std.rational; type Rational = Field<Int>. Long header documents the executable-completeness gap (Field<Int> denotes reciprocal: fn(Int) -> Int which is denotationally undefined for non-units in ℤ) as a P5 tracked scaffold with named dissolution trigger (FieldOfFractions<R> derived-carrier or NonZero/Result reciprocal refinement), consistent with the prior #1370 audit's deferred completeness gap. - src/v3/compiler/src/bootstrap_regen_fresh.rs: rational.dag wired into std_fixtures (between integer.dag and string_type.dag). - src/v3/compiler/tests/integration.rs: parse-corpus extended; new parser smoke handwritten_parser_accepts_rational_dag. - src/v3/compiler/tests/integration/m2_substrate_inhabitance_test.rs: new ratchet rational_resolves_to_field_over_int pinning span (rational.dag), outer Field instantiation, and Int carrier (alias-chain walk through the construction-chain Int alias). - src/v3/compiler/tests/integration/parse_corpus_manifest.txt: refreshed. - bootstrap_*_generated.rs regenerated. Hard boundaries respected: - No NonZero/Result reciprocal semantics in this PR (deferred per dispatch). - No refinement syntax. - No consumer migration. - No tokenizer/literal-grammar work. * chore: apply cargo fmt * revert(std): drop Rational = Field<Int> alias pivot; add FieldOfFractions<R> 6Q audit Director ratified Option 2 substrate-split (inbox #1288 #4362643231) after the same M9 modeling-faithfulness finding that retired AbelianGroup<Nat>: under the standard parametric reading of Field<T>, T is the carrier of field operations including reciprocal: fn(T) -> T. With T = Int, that denotes 'every Int has a multiplicative inverse' — false for ℤ (only ±1 are units). ℚ is the field of fractions derived from ℤ; ℚ's carrier is not Int itself. Reverted in this commit: - dsl/std/rational.dag: deleted (was authoring the false alias). - src/v3/compiler/src/bootstrap_regen_fresh.rs: rational.dag fixture registration removed. - src/v3/compiler/tests/integration.rs: parse-corpus + parser smoke reverted. - src/v3/compiler/tests/integration/m2_substrate_inhabitance_test.rs: rational_resolves_to_field_over_int ratchet removed (per Director's 'remove any ratchet that blesses Field<Int>' guard). - bootstrap_*_generated.rs regenerated to track the revert. - parse_corpus_manifest.txt restored to origin/main state. Added in this commit: - docs/audit/t-numeric-construction-field-of-fractions-6q.md: 6Q substrate-introduction audit for the prerequisite FieldOfFractions<R> algebra surface. Recommends Shape C (opaque atom; carrier-representation deferred to per-target emission) per Director's 'no pair/quotient numerator-denominator representation' boundary and the GroupCompletion precedent (#1448). Once Shape C lands, Slice 4 becomes a one-line edit: type Rational = Field<FieldOfFractions<Int>>. * WIP: tidy-tern-769 * test(v3): add per_call_descent_evidence_is_single_lookup_authority ratchet T-E-P-Producer-Broadening gate (2) — e_p_call_pattern_lookup_authoritative (per Director's path-(1) selection on inbox #1288 #4363830953). Pins per_call_descent_evidence as the single lookup authority for per-call SubValueRelation evidence over TransformTarget::Callable transforms. Behavioral cardinality assertion: - Every Callable transform in the user fixture file appears in exactly one CallDescentEvidence entry (uniqueness; no parallel producer appending duplicates). - The producer's emitted set (filtered to fixture file) equals the enumerated Callable-transform set in the fixture (totality; producer is not silently dropping call sites). - Every entry corresponds to a real Callable-targeted transform (no synthetic entries from a parallel walker). Visibility unchanged: SubValueRelation enum constructors stay pub for existing test usage (m2_substrate_inhabitance_test.rs:577 and the new e_p tests construct SubValueUnknown directly); the producer helpers classify_call_argument and arithmetic_descent_relation are already private to dag.rs. No producer broadening, no substrate carrier introduction, no lens consumer in this slice. Per dispatch (#1288 #4363830953): S-sized ratchet only; gates (1) and (3) tracked as separate slices. * test(v3): expand E-P gate-2 fixture to exercise cross-template branch Reviewer caught: gate (2) single-authority claim must hold for both branches of per_call_descent_evidence's match (caller_template == callee_template self-recursion AND resolved cross-template fail-closed), but the prior fixture only had countdown self-recursion. A future producer that adds a parallel walker for cross-template evidence (rather than going through the existing fail-closed path) would slip past a self-recursion-only ratchet. Restored the two-function fixture (countdown + caller→helper) so the sanity-floor assertion >= 2 Callable transforms exercises both branches, and the coverage-equality assertion enforces single-authority over the cross-template fail-closed path too. Verified locally with RUST_MIN_STACK=33554432: test passes. * polish(test): trim duplicate comments + dead let; align module doc with span-scoped oracle Cursor exploratory feedback (#1514): two duplicate comment blocks merged into one, no-op 'let _ = ();' placeholder dropped (subsumed by neighbour comment), variable renamed owned_callable_transforms → fixture_callable_transforms (span-based, not body-bind-based) for accuracy, module doc bullet rewritten to describe the span-scoped oracle the code actually uses. No behavioral change; tightens local-readability of the ratchet. * fix(v3): reject duplicate fields in expression-position named-variant constructors `lower_record_to_structural` (data body) and `lower_record_literal_expr` (anonymous record expression) already rejected duplicate field labels before type-field projection. `lower_variant_record_expr` — the path for named-constructor record payloads at expression position, e.g. `Ready { code: 1, code: 2, retry: false }` inside an fn body — silently projected via `find()` over the literal's fields, dropping the second occurrence and constructing a normal payload. Mirror the duplicate-detection pattern from `lower_record_literal_expr` (HashSet pre-pass returning a typed ResolveError on collision) so the third record-shape lowering path now also fails closed. Adds `expr_named_variant_duplicate_payload_fields_fail_closed` covering the previously-silent expression-context path; the existing `data_body_named_variant_duplicate_payload_fields_fail_closed` covers the data-body context. Closes the "Duplicate record-literal fields silently accepted" row of docs/debt/r3-debt-paydown-ledger-2026-05-02.md. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * chore: refresh parse_corpus_manifest after induction.dag content change PR #1543 modified `src/v3/std/induction.dag` to correct a SubValueRelation algebra claim but did not refresh the SG-2 parser-corpus snapshot, leaving the committed manifest hash stale relative to the actual file. The v3 suite's `handwritten_parse_snapshot_matches_manifest` reproduced this on the merge into this branch — refreshing via the documented helper. Generated by: cargo test -p v3-compiler --test integration \\ refresh_handwritten_parse_snapshot_manifest -- --ignored Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * chore: refresh bootstrap_generated.rs after induction.dag content change PR #1543 modified `src/v3/std/induction.dag` (SubValueRelation algebra correction) but did not regenerate the committed bootstrap snapshot. v3 CI's `regen_bootstrap --verify` gate flagged span-offset drift in the induction.dag entry on this branch. Regenerated via the documented helper. Generated by: cargo run -p v3-compiler --features bootstrap-regen-fresh \\ --bin regen_bootstrap Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

briansrls added 30 commits April 30, 2026 17:47

WIP: tidy-tern-769

2592b65

chore: apply cargo fmt

3453a17

WIP: tidy-tern-769

c36e956

Merge remote-tracking branch 'origin/main' into session/tidy-tern-769

8d10039

# Conflicts: # src/v3/compiler/src/bootstrap_generated.rs # src/v3/compiler/src/bootstrap_generated_without_parse_surface.rs

WIP: tidy-tern-769

f008b6f

chore: apply cargo fmt

dce2fc0

WIP: tidy-tern-769

0f2d0d0

Merge remote-tracking branch 'origin/main' into session/tidy-tern-769

75772bd

Merge remote-tracking branch 'origin/main' into session/tidy-tern-769

37e9b38

Merge remote-tracking branch 'origin/main' into session/tidy-tern-769

16207e9

Merge remote-tracking branch 'origin/main' into session/tidy-tern-769

962d973

# Conflicts: # src/v3/compiler/src/bootstrap_generated.rs # src/v3/compiler/src/bootstrap_generated_without_parse_surface.rs

Merge remote-tracking branch 'origin/main' into session/tidy-tern-769

46e92f0

Merge remote-tracking branch 'origin/main' into session/tidy-tern-769

9fd943e

# Conflicts: # src/v3/compiler/src/bootstrap_generated.rs # src/v3/compiler/src/bootstrap_generated_without_parse_surface.rs

Merge remote-tracking branch 'origin/main' into session/tidy-tern-769

ed0f59c

Merge remote-tracking branch 'origin/main' into session/tidy-tern-769

361e128

WIP: tidy-tern-769

c806aaa

feat(std): add Magnitude opaque carrier (T-Numeric-Construction Slice 1)

64edc56

Merge remote-tracking branch 'origin/main' into session/tidy-tern-769

6b756a9

WIP: tidy-tern-769

874e321

Merge remote-tracking branch 'origin/main' into session/tidy-tern-769

08bb95a

WIP: tidy-tern-769

d279af0

Merge remote-tracking branch 'origin/main' into session/tidy-tern-769

07cb719

Merge remote-tracking branch 'origin/main' into session/tidy-tern-769

6440ca0

WIP: tidy-tern-769

c4a56fa

Merge remote-tracking branch 'origin/main' into session/tidy-tern-769

83ac68d

Merge remote-tracking branch 'origin/main' into session/tidy-tern-769

09f0af7

Merge remote-tracking branch 'origin/main' into session/tidy-tern-769

00d4c7f

Merge remote-tracking branch 'origin/main' into session/tidy-tern-769

dd23d33

Merge remote-tracking branch 'origin/main' into session/tidy-tern-769

5f533e0

Merge remote-tracking branch 'origin/main' into session/tidy-tern-769

63d7075

briansrls added 13 commits May 2, 2026 01:25

Merge remote-tracking branch 'origin/main' into session/tidy-tern-769

06f570b

Merge remote-tracking branch 'origin/main' into session/tidy-tern-769

4acefa1

Merge remote-tracking branch 'origin/main' into session/tidy-tern-769

48efa09

Merge remote-tracking branch 'origin/main' into session/tidy-tern-769

ce232a5

Merge remote-tracking branch 'origin/main' into session/tidy-tern-769

13e9cd7

Merge remote-tracking branch 'origin/main' into session/tidy-tern-769

8dcd79c

Merge remote-tracking branch 'origin/main' into session/tidy-tern-769

15e5e07

Merge remote-tracking branch 'origin/main' into session/tidy-tern-769

7610ae5

Merge remote-tracking branch 'origin/main' into session/tidy-tern-769

ee80769

Merge remote-tracking branch 'origin/main' into session/tidy-tern-769

ac304d6

Merge remote-tracking branch 'origin/main' into session/tidy-tern-769

8e63fb0

WIP: tidy-tern-769

52985b3

briansrls changed the title ~~tidy-tern-769~~ test(v3): T-E-P-Producer-Broadening gate (2) — single-lookup-authority ratchet May 2, 2026

briansrls marked this pull request as ready for review May 2, 2026 13:15

This was referenced May 2, 2026

session/jolly-ram-908 · jolly-ram-908 #1130

Closed

session/tidy-tern-769 · tidy-tern-769 #1288

Closed

briansrls commented May 2, 2026

View reviewed changes

briansrls merged commit 54b4bc5 into main May 2, 2026
4 checks passed

This was referenced May 2, 2026

session/zesty-dove-500 · zesty-dove-500 #1526

Closed

session/zesty-bear-812 · gunbc Director #828

Closed

briansrls mentioned this pull request May 5, 2026

docs(audit): R3 debt sweep framework draft (Phase 1) — gunbc#828 dispatch #1804

Closed

Navigation Menu

Repository navigation

test(v3): T-E-P-Producer-Broadening gate (2) — single-lookup-authority ratchet - #1514

test(v3): T-E-P-Producer-Broadening gate (2) — single-lookup-authority ratchet#1514
briansrls merged 114 commits into
mainfrom
session/tidy-tern-769

briansrls commented May 2, 2026 •

edited

Loading

Uh oh!

briansrls commented May 2, 2026

Uh oh!

briansrls left a comment

Uh oh!

briansrls May 2, 2026

Uh oh!

briansrls commented May 2, 2026

Uh oh!

briansrls commented May 2, 2026

Uh oh!

briansrls commented May 2, 2026

Uh oh!

briansrls commented May 2, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

briansrls commented May 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

E-P producer surface audit (recap)

What lands

Hard boundaries respected

Files

Test plan

Next slices

Uh oh!

briansrls commented May 2, 2026

Uh oh!

briansrls left a comment

Choose a reason for hiding this comment

Uh oh!

briansrls May 2, 2026

Choose a reason for hiding this comment

Uh oh!

briansrls commented May 2, 2026

Uh oh!

briansrls commented May 2, 2026

Uh oh!

briansrls commented May 2, 2026

Uh oh!

briansrls commented May 2, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

briansrls commented May 2, 2026 •

edited

Loading