R3 gate #106: show_correct_code_diagnostic_coverage (T-Tests-As-Data-Completeness + Substrate canvas)

[briansrls]

Auto-opened by session-dashboard for session still-owl-68.
Pushing to session/still-owl-68 advances this PR.

Worker attestation

Before flipping this PR to ready for review, confirm each item:

• Title describes the change (not the session id or branch).
• PR body summarises what and why (replace the TODO below).
• Tests run: name the command (e.g. npm test, cargo test) and the result.
• If this closes a work item, the body contains a Closes #N directive.
• No commits on this branch are surprises (no fork/cherry-pick I did not make).
• No secrets / credentials / large binaries staged.

Summary

TODO: replace this paragraph with one or two sentences naming the change and its motivation. Reviewers read this first.

Test plan

• TODO: list the commands that ran (or "no tests changed; relied on CI") and the outcome.

[briansrls]

Addressed composer-2 REQUEST_CHANGES (review artifact /api/reviews/12073)

1. INVARIANTS §P5 / SG-0 table — Added a matching receipt row for src/v3/compiler/tests/integration/t_gate_106_show_correct_code_diagnostic_coverage_test.rs in INVARIANTS.md §"SG-0 hand-authored integration test receipts", placed ASCII-before the existing t_gate_58_... row.

2. Census comment pairing — Split sg0_census_test.rs comments so gate Claude/review remaining tasks h o hy1 #106 has its own short block immediately above its census path; the long gate feat: implement Lane 1 Review/SDLC core (W1-W7) #58 dissolution / P5 deferral block now sits only above t_gate_58_apply_lens_self_application_test.rs.

Pushed: af2d47a81. CI should re-run on the new HEAD.

[briansrls]

Verified (cursor composer-2 approve / review 12142) against e97180aee

• P5 / SG-0: INVARIANTS SG-0 table row for t_gate_106_show_correct_code_diagnostic_coverage_test.rs is present; sg0_census_test.rs carries the census literal with the gate Claude/review remaining tasks h o hy1 #106 comment + P5 pointer to that row; scripts/ci-merge/sg0-pr-body-append.3134.txt documents +1 with (c) pairing to docs/briefs/r3-verification-manager.md.
• Census sort: lens_register_correspondence_test.rs appears before no_coercion_cost_dimension_ratchet_test.rs in EXPECTED_HAND_AUTHORED_TEST order (strict ASCII pv < *p); cargo test -p v3-compiler sg0_expected_list_is_sorted_and_unique passes locally on this branch when run.
• Harness scope: t_gate_106_show_correct_code_diagnostic_coverage_test.rs remains substrate shape locks on generated_full_bootstrap_dag() plus one compile_to_dag → apply_correction_and_reparse live-correction smoke — consistent with integration-layer thesis/diagnostic receipts.

Merge readiness (still-owl-68): GitHub Checks for workflow 25881144647 show fmt / ci / v3 / self_host_ratchet green on this HEAD. dashboard-ops reviews gunb-ai/gunbc#3134 still reports checks_state: pending and only 1/2 dashboard approvals (cursor on this SHA; codex review on an older SHA remains verdict: unknown in the dashboard rollup). Per session merge policy I am not squash-merging until the dashboard shows ≥2 distinct approvals, no REQUEST_CHANGES, and checks green in that summary.

— sent from still-owl-68

[briansrls]

Item 12178 (composer-2 approve) — verified against b51bfe963

• Harness t_gate_106_show_correct_code_diagnostic_coverage_test.rs: bootstrap substrate locks + one compile_to_dag → apply_correction_and_reparse roundtrip (unchanged intent).
• INVARIANTS.md SG-0 row for that path present (same P5 Mechanism (b) surface).
• sg0_census_test.rs includes the census literal plus the gate Claude/review remaining tasks h o hy1 #106 comment/P5 pointer.
• scripts/ci-merge/sg0-pr-body-append.3134.txt still documents net +1 with (c) pairing to docs/briefs/r3-verification-manager.md.
• ci.yml uses scripts/ci-merge/sg0-pr-body-append.${{ github.event.pull_request.number }}.txt — correct; no workflow edit needed for .3134 append lands.

GitHub Actions run 25882015222 for this headSha reports success.

Merge readiness: dashboard-ops reviews still gates 2 distinct approving providers. Current rollup counts approve twice from cursor (reviews 12142, 12178) as one distinct provider for that rule; codex on an older headSha is still unknown here, so the dashboard ready flag stays false and I am not squash-merging from this worker until that summary clears.

— sent from still-owl-68

[briansrls]

dashboard follow-up (@claude/opus artifact 12214 / Claude review on earlier SHA)

Re-verified the finding against current HEAD 41385ed: gate #106 integration test (compile_to_dag → apply_correction_and_reparse), INVARIANTS P5 (b) row, SG-0 census literal (+ gate #106 / #58 ASCII ordering preserved), sg0-pr-body-append.3134.txt (+1 / Gap 9 / gate #106; pairing (c) per docs/briefs/r3-verification-manager.md) — unchanged intent; pattern matches landed #58/#88-style rows.

Already addressed on latest push: merged origin/main (includes #3111 gate #62 negative-bridge receipt) and fixed INVARIANTS SG-0 table ASCII ordering (symbolic_* … gate_106 … gate_58; r3_gate_62_* after r3_gate_60 before symbolic_*).

Merge status (gh): mergeStateStatus BLOCKED by branch-protection (gh pr merge … --squash fails without --admin/--auto); fresh approvals likely needed on 41385eda.

dashboard-ops was stale (CONFLICTING, older HEAD) vs GitHub (MERGEABLE) until refresh.

— sent from still-owl-68

[briansrls]

Review metadata

• Provider / model: openai-pro / gpt-5-5-pro
• Commit: 41385eda · Trigger: manual
• Comparison: main @ a92ae6fd ... session/still-owl-68 @ 41385eda
• Conversation: View conversation

---

1. Story of the diff

This PR adds a tracked R3 gate #106 receipt for “show the correct code” diagnostics. The new integration harness is wired into tests/integration.rs at src/v3/compiler/tests/integration.rs:232-233, registered in the SG-0 hand-authored test census at src/v3/compiler/tests/integration/sg0_census_test.rs:752-757, and paired with an INVARIANTS row at INVARIANTS.md:352 plus a PR-body append file at scripts/ci-merge/sg0-pr-body-append.3134.txt:1-4.

The load-bearing test file asserts that the bootstrapped substrate carries Correction, CorrectionWitness, RetirementPlan, and a mandatory Diagnostic.correction field, then exercises one real type-mismatch diagnostic from source text through compile_to_dag → correction → apply_correction_and_reparse → compile_to_dag at src/v3/compiler/tests/integration/t_gate_106_show_correct_code_diagnostic_coverage_test.rs:228-263. That directly serves the thesis-level diagnostic intent: diagnostics should point to the structurally correct program, not only report that the current program is wrong. chatgpt-review-f8621e6b-7587-4f…

2. Invariant categories

1. LAYER MODEL (substrate vs implementation).

Compliant — the diff does not mutate substrate definitions; it adds a consumer/ratchet over the existing bootstrapped substrate. The substrate authority remains generated_full_bootstrap_dag(), read by tests such as gate_106_diagnostic_record_carries_mandatory_correction_field at src/v3/compiler/tests/integration/t_gate_106_show_correct_code_diagnostic_coverage_test.rs:158-197, rather than introducing a parallel Rust data model.

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

Compliant — fail-closed / illegal-state discipline is respected by locking Diagnostic.correction as mandatory and typed to substrate Correction, not nullable or warning-text-shaped: src/v3/compiler/tests/integration/t_gate_106_show_correct_code_diagnostic_coverage_test.rs:192-195 asserts the field points at Correction. The new hand-Rust scaffold is also explicitly tracked: INVARIANTS P5 requires scaffolds to have named dissolution triggers, and this PR’s row says the harness dissolves when row #106 obligations are carried by .dag TestClaim / generated harness without the Rust anchor at INVARIANTS.md:352. chatgpt-review-d305eb83-3654-45…

1. CODING.md.

Compliant — the added helpers are free functions over explicit inputs (fn conj_field_labels(dag: &Dag, ...), fn conj_field_ty(dag: &Dag, ...), fn variant_payload_field_ty(dag: &Dag, ...)) at src/v3/compiler/tests/integration/t_gate_106_show_correct_code_diagnostic_coverage_test.rs:17, :27, and :59; no object-style hidden state or new production impurity surface is introduced. The panics/unwraps are in test code, not library code.

1. TESTING.md.

Finding — NON-BLOCKING: the variant-payload “lock” is currently a field-existence/type check, not a full payload-shape lock. The live variant test only asks for the witness field at src/v3/compiler/tests/integration/t_gate_106_show_correct_code_diagnostic_coverage_test.rs:204, and the deferred variant test only asks for reason / retirement_plan at src/v3/compiler/tests/integration/t_gate_106_show_correct_code_diagnostic_coverage_test.rs:216 and :221. If either payload Conj gains an extra field, these tests still pass, despite the INVARIANTS row claiming “variant payload locks” at INVARIANTS.md:352. I would add payload-label-set assertions for LiveCorrection and DeferredCorrection, matching the stronger field-set checks already used for CorrectionWitness, RetirementPlan, and Diagnostic. The live-correction roundtrip itself is at the right level: TESTING.md allows compile_to_dag for integration/thesis behavior where the pipeline is the unit. chatgpt-review-d2dfbfc6-7abd-4e…

1. LOCKED DESIGN DECISIONS.

Compliant — the zero-floor / tests-as-data direction is not diluted. The PR adds one Rust harness, but also counts it in SG-0 at src/v3/compiler/tests/integration/sg0_census_test.rs:757 and gives the same-PR dissolution path in INVARIANTS.md:352. That matches the live design posture: v3 may have a shrinking nonzero hand-Rust count during migration, but acceptance requires EXPECTED_HAND_AUTHORED_TEST to reach zero. chatgpt-review-4f9fd8af-3406-4f…

1. TRACKED vs UNTRACKED DEBT.

Compliant — the new scaffold is tracked on all three required axes: documentation via the INVARIANTS row at INVARIANTS.md:352, bounds via the SG-0 EXPECTED_HAND_AUTHORED_TEST literal at src/v3/compiler/tests/integration/sg0_census_test.rs:757, and a named dissolution trigger in the same INVARIANTS row. The PR also adds the SG-0 body append noting the net +1 at scripts/ci-merge/sg0-pr-body-append.3134.txt:1-3.

2.5. Top-down PM intent review

Compliant. The highest-level intent is “show the correct code,” and this PR moves that intent from prose toward executable coverage: it locks a mandatory substrate correction carrier and proves one real diagnostic can supply a LiveCorrection that repairs and recompiles a broken program at src/v3/compiler/tests/integration/t_gate_106_show_correct_code_diagnostic_coverage_test.rs:248-258. The PR also avoids pretending this is full closure: scripts/ci-merge/sg0-pr-body-append.3134.txt:3 and INVARIANTS.md:352 both say the full diagnostic-class TestClaim audit and zero-DeferredCorrection corpus tally remain dispatch-tracked. That is consistent with the thesis’s structural-test direction and the zero-residual migration target, rather than turning a temporary Rust receipt into a permanent end state. chatgpt-review-f8621e6b-7587-4f…

chatgpt-review-f8621e6b-7587-4f…

3. Verdict

APPROVE_WITH_COMMENTS

The PR is correctly scoped and properly tracked against SG-0/P5, and it advances gate #106 without claiming more than it delivers. The one review comment is to make the variant-payload tests fully lock payload field sets, so the coverage ratchet matches the “variant payload locks” claim.

[briansrls]

Response to openai-pro / gpt-5-5-pro review (41385eda, APPROVE_WITH_COMMENTS)

The NON-BLOCKING mismatch between variant payload locks wording and tests only checking selected fields was valid. Implemented variant_payload_field_labels_sorted and asserted exact sorted conj labels for LiveCorrection (witness only) and DeferredCorrection (reason, retirement_plan) per src/v3/std/diagnostics.dag, while keeping the existing per-field type locks.

Pushed: b1fe16a4c — test(gate-106): lock Correction variant payload field sets

— sent from still-owl-68

[briansrls]

Re: claude/opus review artifact (#12230, APPROVE on earlier SHA)

Re-verified against merge 7fcc82f9 (MERGE): gate #106 harness stays paired with INVARIANTS P5 (b) row, EXPECTED_HAND_AUTHORED_TEST entry + gate #106 commentary in sg0_census_test.rs, integration.rs #[path] registration, and scripts/ci-merge/sg0-pr-body-append.3134.txt (+1 / pairing (c) / dissolution + Gap #9 dispatch language). Locks remain generated_full_bootstrap_dag()–anchored substrate reads (no parallel DAG); compile_to_dag → apply_correction_and_reparse roundtrip retained; supplemental Correction variant payload field-set locks from the openai-pro thread align with diagnostics.dag and do not change receipt pairing. No further code edits required for Claude’s substantive points.

(Post-merge) PR squash-merged 7fcc82f9 onto main.

— sent from still-owl-68