nimble-pike-489

[briansrls]

Opened from session-dashboard for session nimble-pike-489.

⚠️ Auto-merge of origin/main failed during this auto-push.
The branch has been pushed as-is; resolve the conflict manually before merging:

git -C <worktree> fetch origin main
git -C <worktree> merge origin/main
# resolve conflicts, then commit + push

Conflicted file(s):

• Cargo.toml

[briansrls]

Manager APPROVE. Clean R2-closure-pre-pass scaffold matching the dispatch:

• Sibling crate src/v3/grounding_tests/ with workspace registration ✓
• Stratum A generalized from pilot kernel-integer precedent (grounding_pilot:408-460) to LanguageSpec Phase 1 rows on main (Rust 9 / Python 18 / Go 13) ✓
• RowFingerprint + stratum_a_list_digest produces lockstep-comparable digests; routing-parity test compares bootstrap-loaded vs standalone-compiled to catch fold drift ✓
• Determinism check (stratum_a_determinism_independent_of_list_walk_order) — exact discipline the brief named ✓
• Row-count regression (stratum_a_row_counts_match_director_phase1) — catches silent additions/drops ✓
• Lane-local GroundingTestsDiagnostic with docs(briefs): author T-Ground-Diagnostic lane brief #1216 convergence-pattern citation ✓
• SG-0: consumes v3-compiler APIs; no compiler edits ✓
• Cross-lane: grounding_lifetime_sibling_wires_through_bootstrap regression confirms grounding_lifetime integration matures with bootstrap ✓

Mark ready and merge when CI green. R2 closure-pre-pass complete for both partial-shippable lanes (this + #1271 CrossTarget-Meta).

— sent from silent-ant-322 (inbox #1133); reply at #1133

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: aa2cc236 · Trigger: schedule
• Thinking: 268s wall

---

BLOCKING (4)

Root Cause

• src/v3/grounding_tests/src/stratum_a.rs source authority is encoded once as a crate-relative include and again as a substrate path → make the include path target the same src/v3/std files or derive the source through the existing compiler bootstrap fixture authority.
• src/v3/grounding_tests/src/stratum_a.rs the test relies on internal projection helpers without a test-visible API → move the test into stratum_a or add cfg(test) pub(super) wrappers/constants for the private helpers it exercises.
• src/v3/grounding_tests/src/stratum_a.rs lockstep diagnostics need a structural rendering of RowFingerprint → derive Debug on RowFingerprint or render the existing canonical_line values in the diagnostic detail.
• src/v3/grounding_tests/src/stratum_a.rs graph-shape lookups are modeled as infallible helpers → make list_rows, method_emit_template_variant_label, and placeholder_variant_label return Result and thread the diagnostic through callers.

⚠️ The new verification crate is well scoped, but it currently has compile blockers plus one fail-closed diagnostic gap that should be fixed before landing.

[briansrls]

Verified on current `HEAD` of this branch: `src/v3/grounding_tests/src/stratum_a.rs` no longer uses `include_str!` for the `*_method_template_contracts.dag` files at all (the blocking pattern that could resolve `../std/…` from `grounding_tests/src/` into `src/v3/grounding_tests/std/` is gone). Stratum A verification reads only `generated_full_bootstrap_dag()`, i.e. the compiler-embedded snapshot of `src/v3/std/` authorities.

Rustdoc at the top of `stratum_a.rs` now documents that intentional choice and the `../std` foot-gun so we do not regress.

Prior commit touching this: `335bb9d26` (chore: drop unused `include_str` paths after bootstrap-only verify).

— sent from nimble-pike-489

[briansrls]

Codex schedule review (38f6887) — SingleTemplate arity

Verified: emit_template_canonical only used payload.first() for SingleTemplate, so extra payload fields would be silently ignored while HigherOrderTemplates already enforced payload.len() == 2.

Fix (pushed 3d957af6b): SingleTemplate now fails closed with StratumADagProjectionFailed at step emit_template_canonical.SingleTemplate when payload.len() != 1, then projects payload[0] as a string literal (same pattern as the HOT branch).

cargo test -p v3-grounding-tests --lib and cargo clippy -p v3-grounding-tests --all-targets -- -D warnings pass.

— sent from nimble-pike-489

[briansrls]

Inline @ stratum_a.rs:146 — SingleTemplate arity

Verified on current HEAD: this is already fixed. At emit_template_canonical / SingleTemplate, line ~146 is now an explicit payload.len() != 1 guard returning StratumADagProjectionFailed (emit_template_canonical.SingleTemplate); projection uses payload[0] only after that check (landed 3d957af6b). The bot text matches the pre-fix shape (payload.first() with no arity check).

No further code change needed for this thread.

— sent from nimble-pike-489

[briansrls]

Codex schedule review (c01517b6) — APPROVE

Re-verified on current HEAD (3d957af6b, includes follow-up emit_template_canonical SingleTemplate arity fail-closed after that snapshot): there are no actionable findings in that review—verdict is approval only.

cargo test -p v3-grounding-tests passes (6 tests). No fix commit required for this item.

— sent from nimble-pike-489

[briansrls]

Review metadata

• Provider / model: claude / claude-opus-4-7
• Commit: 3d957af6 · Trigger: schedule
• Comparison: origin/main @ 68c1b9cc ... review/pr-1274-3d957af6 @ 3d957af6
• Thinking: 34s wall

---

Verdict: APPROVE — new sibling crate v3-grounding-tests is consumer-only (SG-0 honored), all paths fail-closed via GroundingTestsDiagnostic, the lane-local Result carrier carries an explicit Practice-4 🟡 ledger entry with a named dissolution trigger (manager #1216 / EmissionDiagnostic), so the scaffold is tracked debt rather than parallel substrate. Lockstep verification re-runs generated_full_bootstrap_dag() and digests via BTreeMap for order-invariance — determinism is asserted by a forward-vs-reverse test. Row counts are Director-locked constants with a clear update path comment.

Exploratory observations (non-blocking):

• assert_expected_row_count (stratum_a.rs:~395) silently treats an unknown list_name as expected = 0 via unwrap_or(0). Today the only callers pass the three module-level constants, but if this is ever called with a typo it would surface as a confusing "expected 0, got N" RowCountMismatch rather than a distinct "unknown list" projection failure. Consider a StratumADagProjectionFailed for the missing-entry case.
• verify_stratum_a_lockstep_all_targets constructs two full bootstrap Dags per call and is invoked from a #[test]; fine for now, but if more Stratum tests pile on this crate, a shared OnceLock<Dag> would keep test time linear.

[briansrls]

Claude schedule review (3d957af6) — exploratory follow-ups

1. assert_expected_row_count / unknown list_name: Valid. Replaced unwrap_or(0) with StratumADagProjectionFailed at step assert_expected_row_count.unknown_list and a detail line listing known Director-locked lists. Added assert_expected_row_count_unknown_list_is_not_silent_zero_expected. Pushed af11a3b1e.

2. OnceLock<Dag> for repeated bootstrap: Leaving as-is for now — still only one lockstep integration test plus cheap unit tests; double generated_full_bootstrap_dag() keeps the witness explicit. Revisit if Stratum tests multiply enough that CI time matters.

cargo test -p v3-grounding-tests and cargo clippy -p v3-grounding-tests --all-targets -- -D warnings pass on af11a3b1e.

— sent from nimble-pike-489

[briansrls]

Review metadata

• Provider / model: codex / codex-default
• Commit: af11a3b1 · Trigger: schedule
• Comparison: origin/main @ 9e696d3d ... review/pr-1274-af11a3b1 @ af11a3b1
• Thinking: 327s wall

---

Verdict: APPROVE

This looks clean. The diff is a new implementation/test scaffold crate reading the generated bootstrap DAG, with typed fail-closed diagnostics and a documented yellow scaffold path for the local diagnostic enum. I don’t see concrete violations of the pinned invariants, modeling discipline, coding guidance, or testing guidance.

Verified with cargo test -p v3-grounding-tests — 7 passed.

[briansrls]

Codex schedule review (af11a3b1) — APPROVE

Re-checked on HEAD (af11a3b1e): verdict is approval only; no blocking or change-request items to implement.

cargo test -p v3-grounding-tests — 7 passed (matches review).

— sent from nimble-pike-489

[briansrls]

Review metadata

• Provider / model: cursor / composer-2
• Commit: 9c489fe9 · Trigger: schedule
• Comparison: origin/main @ 9e696d3d ... review/pr-1274-9c489fe9 @ 9c489fe9
• Thinking: 115s wall

---

Findings: None. This diff only adds the v3-grounding-tests crate plus workspace wiring. It’s implementation code (test/verification harness), not new substrate on the Dag. Projection paths return Result<_, GroundingTestsDiagnostic> with explicit variants (src/v3/grounding_tests/src/stratum_a.rs: mismatches, missing fields, unknown lists like 379:397:src/v3/grounding_tests/src/stratum_a.rs), which matches P3 / modeling-discipline Practice 1 (fail-closed). Stratum A reads generated_full_bootstrap_dag() only and documents why it avoids include_str! relative paths (7:11:src/v3/grounding_tests/src/stratum_a.rs), which matches P2 / “facts flow forward” / single structural read for those rows. GroundingTestsDiagnostic is annotated Practice 4 🟡 with a named hand-off (docs/briefs/t-ground-diagnostic.md / #1216) in 15:17:src/v3/grounding_tests/src/diagnostic.rs, satisfying the scaffold ledger/trigger expectation for a new multi-variant diagnostic enum. Style matches CODING.md (data + free functions, structured errors). Tests target bootstrap determinism and parity; using full bootstrap here is consistent with TESTING.md (“when compile/bootstrap is the unit”). I did not identify any diff line that violates the pinned rubric.

Verdict: APPROVE — Scoped addition, boundaries and fail-closed behavior are clear, and the YELLOW diagnostic classification is documented with a dissolution pointer.

Exploratory (optional): INVARIANTS.md P5 “per-PR gate” for new hand-Rust under src/v3/ refers to the PR description receipt; that isn’t in the diff, so confirm on GitHub that the template gate is satisfied if your process treats it as merge-blocking.

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: af11a3b1 · Trigger: schedule
• Thinking: 269s wall

---

BLOCKING (2)

Root Cause

• src/v3/grounding_tests/src/stratum_a.rs MethodRef resolution treats a name-shaped record as the method registry authority -> resolve the MethodDeclaration declaration id once and require the referenced declaration's declared type or meta_tag to match before extracting name.
• src/v3/grounding_tests/src/stratum_a.rs PlaceholderConvention projection is label-only while MethodEmitTemplate projection validates payload shape -> add a canonicalizer that asserts zero payload for nullary Phase 1 variants or validates and includes each payload field before constructing RowFingerprint.

⚠️ Two fail-closed gaps remain in the new Stratum A projection.

[briansrls]

BLOCKING: method_registry_name reads the referenced declaration's name field without proving the declaration is a MethodDeclaration, so a MethodRef to any record with a string name can pass the routing check (Every Dependency Is A Substrate Fact / API-level enforcement).

[briansrls]

BLOCKING: placeholder_convention ignores the variant payload, so malformed or future payload-bearing PlaceholderConvention rows collapse to the same fingerprint instead of failing closed (Fail-closed / Facts flow forward).

[briansrls]

Review metadata

• Provider / model: claude / claude-opus-4-7
• Commit: 7490ebfa · Trigger: schedule
• Comparison: origin/main @ b9504ab2 ... review/pr-1274-7490ebfa @ 7490ebfa
• Thinking: 40s wall

---

Verdict: APPROVE — clean, narrowly-scoped new test crate (v3-grounding-tests) with no edits to src/v3/compiler/ (SG-0 respected). Diagnostics are fail-closed Result-returning (no silent defaults — see the explicit assert_expected_row_count_unknown_list_is_not_silent_zero_expected regression). Practice 4 🟡 ledger is documented on GroundingTestsDiagnostic with a named dissolution trigger (EmissionDiagnostic / #1216), satisfying the tracked-bridge bar. Lockstep digest is computed from two independent generated_full_bootstrap_dag() constructions, which is the right shape for routing-stability parity. No findings.

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: 7490ebfa · Trigger: schedule
• Thinking: 305s wall

---

BLOCKING (1)

Root Cause

• src/v3/grounding_tests/src/stratum_a.rs MethodTemplateContract and MethodRef are projected through open Vec field scans instead of closed schema canonicalizers → introduce typed schema readers that reject duplicate/unknown fields and either fingerprint every known fact or document explicit discards.

Non-blocking — Strengths

• src/v3/grounding_tests/src/diagnostic.rs GroundingTestsDiagnostic has a bounded yellow scaffold receipt naming the EmissionDiagnostic hand-off, so the lane-local carrier is tracked.

⚠️ One new fail-closed projection gap remains in the Stratum A record reader.

[briansrls]

BLOCKING: row_field returns the first matching label and leaves duplicate or extra fields unexamined, so MethodTemplateContract and MethodRef records can carry unconsumed substrate facts without a diagnostic (Facts flow forward / fail-closed).

[briansrls]

Review metadata

• Provider / model: openai-pro / gpt-5-5-pro
• Commit: af11a3b1 · Trigger: manual
• Conversation: View conversation

---

1. Story of the diff

This PR adds a new workspace crate, v3-grounding-tests, and wires it into the root workspace at Cargo.toml:24-25 and Cargo.lock:717-723. The crate is a Stratum A verification scaffold for MethodTemplateContract rows: it reads the compiler’s embedded full bootstrap Dag as the single authority, projects each target’s *_method_template_contracts list into deterministic row fingerprints, checks Director-locked row counts, resolves each dag_method through the method registry, rejects duplicate method keys, and verifies digest determinism across two bootstrap constructions (src/v3/grounding_tests/src/stratum_a.rs:7-11, src/v3/grounding_tests/src/stratum_a.rs:344-440). It also introduces a lane-local typed diagnostic enum with an explicit yellow-scaffold dissolution path toward shared EmissionDiagnostic (src/v3/grounding_tests/src/diagnostic.rs:1-14) and adds focused unit tests for the public verification path, unknown-list fail-closed behavior, digest determinism, and sibling-crate wiring (src/v3/grounding_tests/src/stratum_a.rs:449-507, src/v3/grounding_tests/src/lib.rs:30-55).

2. Invariant categories

1. LAYER MODEL — Compliant. This is implementation/test-lane code, not a substrate mutation: the new crate explicitly consumes v3_compiler::generated_full_bootstrap_dag and avoids parallel include_str! source reads, preserving the embedded bootstrap Dag as the single read authority (src/v3/grounding_tests/src/stratum_a.rs:7-11). No Dag, substrate .dag, or cross-pass substrate type is changed.
2. INVARIANTS.md + modeling-discipline.md — Compliant. Fail-closed and single-authority discipline are handled directly: missing declarations, missing value bodies, wrong list shapes, unknown constructors, malformed rows, missing fields, wrong literal types, duplicate method keys, and row-count drift all return GroundingTestsDiagnostic instead of fabricating defaults (src/v3/grounding_tests/src/stratum_a.rs:31-49, src/v3/grounding_tests/src/stratum_a.rs:70-75, src/v3/grounding_tests/src/stratum_a.rs:235-245, src/v3/grounding_tests/src/stratum_a.rs:352-357, src/v3/grounding_tests/src/stratum_a.rs:383-405). The new coproduct carrier is also explicitly classified as a yellow scaffold with a named dissolution target (src/v3/grounding_tests/src/diagnostic.rs:11-14).
3. CODING.md — Finding, NON-BLOCKING. src/v3/grounding_tests/Cargo.toml:14 adds v3-grounding-lifetime as a normal dependency, but the diff only uses it inside #[cfg(test)] code (src/v3/grounding_tests/src/lib.rs:21-26). Under the explicit-dependencies principle, that edge should be a dev-dependency unless the library surface itself needs v3-grounding-lifetime; as written, consumers of v3-grounding-tests inherit a dependency that appears to be test-only.
4. TESTING.md — Compliant. The PR adds behavior-focused tests at the same layer as the new crate: unknown list names fail closed instead of defaulting to zero (src/v3/grounding_tests/src/stratum_a.rs:470-481), the full Stratum A verification path is exercised through the public API (src/v3/grounding_tests/src/stratum_a.rs:484-487), Director row counts are pinned (src/v3/grounding_tests/src/stratum_a.rs:489-497), and digest idempotence is tested (src/v3/grounding_tests/src/stratum_a.rs:501-506). The tests use the bootstrap Dag because the behavior under test is specifically the embedded row authority, not an incidental pipeline compile.
5. LOCKED DESIGN DECISIONS — Compliant. The PR references the Director-locked Phase 1 row-count contract and encodes it as an assertion target (src/v3/grounding_tests/src/stratum_a.rs:23-25) without altering the underlying locked design or moving the row authority out of the generated bootstrap Dag.
6. TRACKED vs UNTRACKED DEBT — Compliant. The new scaffold is documented, bounded, and has a named dissolution trigger: the diagnostic module states that this is a lane-local Stratum A carrier, keeps the sum small, and names the future handoff to shared EmissionDiagnostic via t-ground-diagnostic.md / manager #1216 (src/v3/grounding_tests/src/diagnostic.rs:1-14). I did not see untracked TODOs, compatibility shims, or temporary parallel source paths in the diff.

3. Verdict

APPROVE_WITH_COMMENTS — The invariant shape is solid: this is a consumer-only verification crate with typed fail-closed diagnostics, no substrate mutation, and an explicit scaffold dissolution path. The only issue I would clean up is the dependency edge: move v3-grounding-lifetime to dev-dependencies unless the non-test library API intentionally depends on it.

[briansrls]

Review metadata

• Provider / model: openai-pro / gpt-5-5-pro
• Commit: 9c489fe9 · Trigger: manual
• Conversation: View conversation

---

1. Story of the diff

This PR adds a new workspace member, src/v3/grounding_tests, to host T-Ground-Tests Stratum A routing-stability checks (Cargo.toml:24-25, src/v3/grounding_tests/Cargo.toml:7-14). The new crate intentionally stays outside src/v3/compiler/ and treats v3_compiler::generated_full_bootstrap_dag() as the single source for *_method_template_contracts rows, rather than include_str!-reading source files from a parallel path (src/v3/grounding_tests/src/lib.rs:7, src/v3/grounding_tests/src/stratum_a.rs:7-11).

Internally, it defines a lane-local typed diagnostic carrier, then walks the bootstrap DAG to resolve each MethodTemplateContract row into a deterministic RowFingerprint, keyed by method name in a BTreeMap, with row-count ratchets for the Director-locked Phase 1 shape (src/v3/grounding_tests/src/diagnostic.rs:14-34, src/v3/grounding_tests/src/stratum_a.rs:23-25, src/v3/grounding_tests/src/stratum_a.rs:344-377). The test module exercises determinism, fail-closed unknown-list handling, locked row counts, and full bootstrap verification (src/v3/grounding_tests/src/stratum_a.rs:449-507).

2. Invariant categories

1. LAYER MODEL (substrate vs implementation).

Compliant — this is an implementation-only verifier crate: it declares “no src/v3/compiler/ edits” at src/v3/grounding_tests/src/lib.rs:7 and reads the embedded bootstrap DAG as the structural authority at src/v3/grounding_tests/src/stratum_a.rs:7-11, without adding substrate types, Dag fields, or new Dag mutation paths.

1. INVARIANTS.md + modeling-discipline.md.

Finding (NON-BLOCKING, illegal states unrepresentable / fail-closed diagnostic shape) — src/v3/grounding_tests/src/stratum_a.rs:433 sets method_name: "<bootstrap-run>".to_string(), for a list-level bootstrap digest mismatch. The diagnostic variant structurally says there is a method name, and its display renders “method {method_name}” (src/v3/grounding_tests/src/diagnostic.rs:56-63), so this path has to fabricate a sentinel method to represent a non-method-scoped failure. A separate list-level variant, or a typed scope field, would keep the diagnostic carrier honest without changing the verifier logic.

1. CODING.md.

Compliant — the core logic is data plus free functions, not hidden-state objects: row projection is decomposed into small helpers such as row_record, row_field, method_name_from_dag_method, and row_fingerprint (src/v3/grounding_tests/src/stratum_a.rs:214-342), and public APIs return Result<..., GroundingTestsDiagnostic> rather than primitive/sentinel outcomes (src/v3/grounding_tests/src/stratum_a.rs:371-377, src/v3/grounding_tests/src/stratum_a.rs:414-440).

1. TESTING.md.

Compliant — the added tests are scoped to the verifier’s actual contract: deterministic digest ordering (src/v3/grounding_tests/src/stratum_a.rs:449-467), fail-closed handling for unknown expected-count rows (src/v3/grounding_tests/src/stratum_a.rs:470-481), locked row-count drift (src/v3/grounding_tests/src/stratum_a.rs:489-498), and the full Stratum A bootstrap verification entry point (src/v3/grounding_tests/src/stratum_a.rs:484-486). Using generated_full_bootstrap_dag() is appropriate here because the embedded bootstrap snapshot is the interface under test.

1. LOCKED DESIGN DECISIONS.

Compliant — the diff does not alter locked design docs; it consumes the Director-locked Phase 1 row counts as an explicit ratchet via EXPECTED_STRATUM_A_ROW_COUNTS (src/v3/grounding_tests/src/stratum_a.rs:23-25) and fails closed when actual counts differ (src/v3/grounding_tests/src/stratum_a.rs:398-404).

1. TRACKED vs UNTRACKED DEBT.

Compliant — the lane-local diagnostic enum is documented as a scaffold, bounded to T-Ground-Tests Stratum A, and names its dissolution trigger: convergence into shared EmissionDiagnostic under t-ground-diagnostic.md / manager #1216 (src/v3/grounding_tests/src/diagnostic.rs:1-5, src/v3/grounding_tests/src/diagnostic.rs:9-12). No unbounded TODOs, hidden bridges, or new temporary include-path side channels are introduced.

3. Verdict

APPROVE_WITH_COMMENTS

The PR is structurally clean as an implementation-only verification crate and handles the important boundary/fail-closed pieces correctly. The one comment is about tightening the lane-local diagnostic shape so list-level bootstrap mismatches do not need a fabricated method-name sentinel.

[briansrls]

gpt-5-5-pro review (af11a3b1) — APPROVE_WITH_COMMENTS / dev-deps

Verified: v3_grounding_lifetime is only imported under #[cfg(test)] in src/v3/grounding_tests/src/lib.rs, so the finding is valid.

Pushed on refactor/grounding-lifetime-dag-boundary-extract: v3-grounding-lifetime moved from [dependencies] to [dev-dependencies] in src/v3/grounding_tests/Cargo.toml (0faa7aef9). Rest of the verdict (consumer-only crate, fail-closed diagnostics, Practice 4 ledger) matches current main-based branch state; no further code changes for those bullets.

cargo test -p v3-grounding-tests and cargo clippy -p v3-grounding-tests --all-targets -- -D warnings pass after the manifest change.

— sent from nimble-pike-489

[briansrls]

gpt-5-5-pro review (9c489fe9) — list-level lockstep diagnostic

Verified: the non-blocking note about method_name: \"<bootstrap-run>\" for list digest drift was valid (StratumALockstepMismatch implied a per-method scope in Display).

Change (pushed b7c2675f5): Replaced with StratumALockstepListDigestMismatch { list_name, detail } and updated verify_stratum_a_lockstep_all_targets to return that variant. Display is now list-scoped (lockstep list-digest mismatch for \{list_name}`). Extended stratum_a_diagnostic_display_is_non_empty` to cover the new variant.

Verdict bullets (layer model, CODING.md, TESTING.md, locked design, tracked debt) remain accurate; no further edits for those.

— sent from nimble-pike-489