Runtime/TestClaim Manager — authority over T-22 eval execution, T-38 structured TestClaimRun verdicts, emitted-code run harness, falsification verdict receipts per PR #3938 §11.1. Critical path for L4 (emit runs and matches .dag eval) and rungs 4-6 on every fixture. First artifact: with Compiler Spi

[briansrls]

Auto-opened by session-dashboard for session neat-hawk-413.
Pushing to session/neat-hawk-413 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]

Review metadata

• Provider / model: openai-pro / gpt-5-5-pro
• Commit: 9117947b · Trigger: manual
• Comparison: main @ 92675275 ... session/neat-hawk-413 @ 9117947b
• Conversation: View conversation

---

1. Story of the diff

This uploaded diff is a narrow receipt-hardening PR: it changes only src/v3/compiler/tests/integration/v4_workflow_ci_runner_dag_smoke_test.rs, not the Runtime/TestClaim Manager implementation itself. The test now requires the generated T-38 manual/generated corpus-eval script to expose explicit receipt probes for return and inverted zero comparisons, and it tightens the accepted tally.fail == Zero && tally.deferred == Zero shape so the conjunction must be the terminal body expression rather than merely appearing somewhere earlier in the script at src/v3/compiler/tests/integration/v4_workflow_ci_runner_dag_smoke_test.rs:1585-1595. chatgpt-review-a3226cfb-5421-44…

The second hunk turns the previous loose regex smoke into a small receipt classifier, receipt_accepts, that normalizes whitespace, rejects explicit return, rejects negated / boolean-inverted zero predicates, and requires the final fail/deferred-zero conjunction at src/v3/compiler/tests/integration/v4_workflow_ci_runner_dag_smoke_test.rs:1616-1632. It then adds positive examples for direct final conjunctions and negative examples for inverted comparisons, NotZero, non-final conjunctions, and explicit-return traps at src/v3/compiler/tests/integration/v4_workflow_ci_runner_dag_smoke_test.rs:1634-1673. chatgpt-review-a3226cfb-5421-44…

2. Invariant categories

1. LAYER MODEL

N/A — the diff is Rust integration-test receipt code only. It does not introduce or mutate Dag substrate types, dag.rs shapes, .dag substrate declarations, L1 behavior variants, type connectives, or cross-pass substrate facts; the changed surface is limited to receipt probes and an embedded Python regex self-test in v4_workflow_ci_runner_dag_smoke_test.rs:1585-1673.

2. INVARIANTS.md + modeling-discipline.md

Compliant — fail-closed / early detection is strengthened. The added explicit_return and inverted_zero_comparison probes at src/v3/compiler/tests/integration/v4_workflow_ci_runner_dag_smoke_test.rs:1585-1589, plus the tail-anchored fail_deferred_conjunction at :1591-1595 and mirrored classifier at :1616-1632, make the receipt reject plausible-but-wrong evidence instead of accepting any loose substring that looks like the corpus tally gate. That matches the P3/C-8 direction: malformed or misleading evidence should fail closed rather than fabricate a plausible pass. chatgpt-review-cc713339-bf2d-4e…

No predicate/walker dissolution issue is introduced: the new boolean helper is inside an integration-test fixture over strings, not a substrate/std/ coproduct predicate or reusable algebraic helper.

3. CODING.md

Compliant — the new helper names describe their contract without explanatory comments: explicit_return, inverted_zero_comparison, fail_deferred_conjunction, normalized_source, and receipt_accepts are small, data-in/data-out test helpers at src/v3/compiler/tests/integration/v4_workflow_ci_runner_dag_smoke_test.rs:1616-1632. This fits CODING.md’s preference for clear names, small functions, structured control flow, and pure helper logic rather than hidden state or method-heavy objects. chatgpt-review-e97592d0-1edd-46…

4. TESTING.md

Compliant — the PR adds behavior-driven regression coverage for the actual receipt contract: direct final corpus tally conjunctions are accepted, while inverted, non-final, and explicit-return cases are rejected at src/v3/compiler/tests/integration/v4_workflow_ci_runner_dag_smoke_test.rs:1634-1673. The test remains a Rust integration receipt, but TESTING.md still allows Rust integration tests during the migration where they are the cleanest shape, while naming .dag TestClaim declarations as the long-term target. chatgpt-review-b63f4ad0-c272-4f…

5. LOCKED DESIGN DECISIONS

N/A — the diff does not edit locked design docs, substrate shape, CI workflow authority files, workflow/ci.dag, or the Runtime/TestClaim Manager contract model. It only tightens an existing T-38 smoke receipt in a Rust integration test.

6. TRACKED vs UNTRACKED DEBT

N/A — no new TODO, scaffold, bridge, compatibility shim, temporary table, or deferred migration marker is introduced. The regex checks are additions to an existing receipt test, and they narrow the accepted evidence instead of adding a parallel steady-state authority.

2.5. Top-down PM intent review

Compliant — the highest-level PM intent relevant here is that Tier 3/L4 verification must establish that emitted code executes and matches .dag evaluation, while tests are on the path toward structural TestClaim data rather than hand-maintained behavior assertions. chatgpt-review-17e4c0d5-4b83-4c…

This PR does not dilute that target: it makes the existing Rust receipt harder to game by requiring the final body to be the direct fail == Zero && deferred == Zero verdict and by rejecting inverted or non-returned lookalikes at src/v3/compiler/tests/integration/v4_workflow_ci_runner_dag_smoke_test.rs:1616-1673. chatgpt-review-a3226cfb-5421-44…

From a PM view, this preserves the intended T-38 direction: workers following the receipt can no longer satisfy it by merely emitting a tally check somewhere while returning a falsified verdict later. It is still a transitional Rust-side receipt, but it reduces false acceptance pressure rather than making the TestClaim/structured-verdict goal optional.

3. Verdict

APPROVE

I found no blocking findings. The diff is test-only, narrows an existing receipt, and strengthens fail-closed coverage for inverted and non-terminal corpus tally verdicts without introducing substrate changes, new scaffolding, or locked-design drift.

[briansrls]

Closeout disposition (Runtime/TestClaim Manager / tidy-crab-578):

T-38 lane: Fully landed on main via #4191 (and prior tranches #4142, #4159, #4180). The inverted-tally receipt regex + falsification smoke on v4_workflow_ci_runner_dag_smoke_test.rs is already on origin/main.

This PR's only file-level delta vs main is Wave1/Wave3 CI honesty-ledger smoke redirects (ci-required-surface-cut-2026-06-01.md), which is CI Manager scope and references a ledger file that does not exist on main — would fail if merged as-is.

No further Runtime/TestClaim action on this branch. Follow-on runtime work remains on open lanes: #4150 (P5 bootstrap-evaluator corpus runtime), #4182 (TestClaimRun E0107 emit preservation).