R3 Cluster M #84 runner-shim SG-0 slice

[briansrls]

Summary

• retire the obsolete t_pb_b_brief_d_fixture_smoke_test.rs host shim
• keep the Brief D coverage on the existing runner-backed .dag suites in t_pb_b_1_dag_runner_test
• remove the deleted shim from the SG-0 hand-authored test census and update nearby docs

SG-0 hand-path delta: -1

Validation

• cargo fmt --check
• git diff --check
• cargo test -p v3-compiler --test integration sg0_census_test -- --nocapture
• cargo test -p v3-compiler --test integration t_pb_b_1_pipeline_smoke_suite_passes_through_runner -- --nocapture
• cargo test -p v3-compiler --test integration t_pb_b_1_contract_diagnostic_smoke_suite_passes_through_runner -- --nocapture
• cargo test -p v3-compiler --test integration t_pb_b_1_contract_port_cost_suite_passes_through_runner -- --nocapture
• cargo test -p v3-compiler --test integration r3_tests_as_data_demonstration_suite_passes_through_runner -- --nocapture

Note

• A broader t_pb_b_1_dag_runner_test filter was also run; it reached the pre-existing ExecuteCommand helper setup case and failed because gunbc_execute_command_bootstrap was not built/discoverable in that invocation. The non-ExecuteCommand suites relevant to this deletion are listed above and passed.

[briansrls]

[Director conformance read — would-approve | clean runner-shim slice]

Small focused slice (+10/−63, 10 files): retires obsolete t_pb_b_brief_d_fixture_smoke_test.rs (−50 LOC) since Brief D coverage is already carried by the existing runner-backed .dag suites in t_pb_b_1_dag_runner_test. Updates SG-0 census + nearby docs.

Conformance:

• INVARIANTS: Net deletion (-1 SG-0 hand-test). Correct direction for R3 close gates Add node override support for non-transport I/O node mocking #8 + Implement SDLC pipeline: worker dispatch, stage handlers, and integration tests #84.
• MODELING: Removes parallel-representation debt (per feedback_parallel_representation_debt) — .dag suite + Rust shim were dual-coverage of same surface; deleting the shim consolidates authority.
• TESTING: Behavioral coverage preserved on the runner-backed .dag suite path; honest scope statement in PR body.
• CODING: No code additions beyond doc/registration updates.

Velocity tripwire: −1 dissolution event. Combined with #2715 (−3), this tick's Phase 3 batch is −4 — already exceeds prior 7-day dissolution total of 3.

PR body discipline: this one IS filled (Summary + SG-0 hand-path delta + Validation commands cited). Cleaner than #2715's auto-template body.

No INVARIANTS violations. Would approve.

(Director read posted as comment; dashboard tally unchanged.)

— sent from zesty-bear-812 (inbox #828); reply at #828

[briansrls]

Review metadata

• Provider / model: openai-pro / gpt-5-5-pro
• Commit: 95172db7 · Trigger: manual
• Comparison: main @ eed86ffc ... session/deep-fox-583 @ 95172db7
• Conversation: View conversation

---

1. Story of the diff

This PR retires the Brief D compile-smoke Rust shim from the SG-0 test slice and reclassifies that path as historical, not active authority. The load-bearing move is deleting src/v3/compiler/tests/integration/t_pb_b_brief_d_fixture_smoke_test.rs, removing its module registration from src/v3/compiler/tests/integration.rs:224-225, and deleting the path from EXPECTED_HAND_AUTHORED_TEST at src/v3/compiler/tests/integration/sg0_census_test.rs:638. The surrounding docs are updated to say the former smoke check is retired in favor of the runner-backed .dag siblings through t_pb_b_1_dag_runner_test, while the Brief D .v3 fixtures remain only historical draft copies with a named future dissolution trigger (docs/briefs/t-pb-b-brief-d.md:9, docs/briefs/t-pb-b-brief-d.md:11). This is a ratchet/shim-retirement PR, not a substrate or behavior change.

2. Invariant categories

1. LAYER MODEL — N/A. The diff does not touch substrate types, Dag storage, cross-pass facts, or new variants; the Rust changes are deletion of an integration-test registration (src/v3/compiler/tests/integration.rs:224-225), deletion of a census entry (src/v3/compiler/tests/integration/sg0_census_test.rs:638), and deletion of the compile-only smoke test file.
2. INVARIANTS.md + modeling-discipline.md — Compliant. P5 / Progress Is Dissolution is handled directly: the old host shim is removed, and the remaining duplicate .v3 fixture copies are explicitly bounded as “historical draft copies only” with a named dissolution trigger: “remove or shrink the .v3 copies once Testgen accepts src/v3/compiler/tests/dag/t_pb_b_1_*.dag as the single maintained source” (docs/briefs/t-pb-b-brief-d.md:11). That matches the invariant that scaffolds need explicit dissolution triggers. chatgpt-review-014178fc-ecac-46…
3. CODING.md — Compliant. No new Rust implementation surface is added. The PR removes the local helper assert_lowers_without_diagnostics and its compile_to_dag / panic! test harness path (src/v3/compiler/tests/integration/t_pb_b_brief_d_fixture_smoke_test.rs:15-26) rather than adding new methods, hidden state, builders, or object-shaped APIs.
4. TESTING.md — Compliant. The deleted Rust test was explicitly a “Lowering proxy only” and not semantic predicate evaluation (src/v3/compiler/tests/integration/t_pb_b_brief_d_fixture_smoke_test.rs:7-10). The docs now point active coverage at the runner-backed .dag suites: docs/briefs/r1-selfhosting-manager.md:182-183, docs/briefs/t-pb-b-1.md:10, and src/v3/compiler/tests/dag/README.md:11 all distinguish retired compile-smoke from TestRunner predicate execution. That is consistent with the project’s .dag-native test trajectory and 0-residual direction.
5. LOCKED DESIGN DECISIONS — Compliant. The PR does not alter a locked design authority. It preserves the live Pure Bootstrap / tests-as-data meaning by shrinking hand-authored Rust test surface rather than adding a new handwritten replacement; the R3 plan update only changes gate chore: package session changes for Lane 2 initiation (L2-0) #74 wording to say the Rust integration test is retired (docs/r3-program-plan.md:300). The relevant live authority is the 0-hand-authored-files / tests-as-data direction, not a contrary locked carve-out.
6. TRACKED vs UNTRACKED DEBT — Compliant. The remaining Brief D .v3 duplicates are tracked, not silently permanent: the diff names the scope (“same three claims”), bounds their role (“historical draft copies only” with no dedicated host-side compile-smoke shim), and gives a dissolution trigger tied to Testgen accepting the .dag siblings as the single maintained source (docs/briefs/t-pb-b-brief-d.md:11). I do not see new TODOs, scaffolds, or temporary shapes introduced by this diff.

2.5. Top-down PM intent review

Compliant. The PM-level intent is to move tests into .dag TestClaim data and drive the hand-authored Rust test surface toward zero; the live design says the goal is “zero hand-authored files in v3’s source tree,” and the thesis/test authorities describe tests as .dag TestClaim declarations with no Rust-side residual under the 0-floor path. This PR preserves that intent: it deletes the Brief D host shim from the integration harness (src/v3/compiler/tests/integration.rs:224-225), removes it from the SG-0 expected hand-authored test census (src/v3/compiler/tests/integration/sg0_census_test.rs:638), and updates the docs to route active meaning through runner-backed .dag siblings rather than a compile-only smoke proxy (docs/briefs/t-pb-b-brief-d.md:9, docs/briefs/t-pb-b-brief-d.md:11). I do not see a must-have target becoming optional, a dissolution goal becoming permanent, or a new hand-written implementation replacing .dag authority.

3. Verdict

APPROVE. This is a clean SG-0 ratchet slice: it removes a compile-only Rust shim, updates the census, and documents the remaining historical duplicate fixtures with a bounded dissolution trigger. No substrate, fail-closed, locked-design, or untracked-debt issue is visible in the diff.

[briansrls]

Review metadata

• Provider / model: openai-pro / gpt-5-5-thinking
• Commit: 2134e9f9 · Trigger: manual
• Comparison: main @ e3ad91d1 ... session/deep-fox-583 @ 2134e9f9
• Conversation: View conversation

---

1. Story of the diff

This PR retires the Brief D compile-smoke Rust shim and updates the surrounding plan/brief language so the source of truth is the runner-backed .dag test path instead of a host-side compile-only proxy. The mechanical change removes t_pb_b_brief_d_fixture_smoke_test from the integration test module list at src/v3/compiler/tests/integration.rs:226, removes it from the SG-0 hand-authored test census at src/v3/compiler/tests/integration/sg0_census_test.rs:646, and deletes the shim file itself. The docs are updated to make the semantic distinction explicit: Brief D .v3 copies are historical draft fixtures, while the landed .dag siblings are evaluated through t_pb_b_1_dag_runner_test, including real predicate execution rather than merely lowering the wrapper module cleanly (docs/briefs/t-pb-b-brief-d.md:7, docs/briefs/t-pb-b-brief-d.md:11, src/v3/compiler/tests/dag/README.md:11). The R3 gate #74 comments are adjusted to say the Rust target being ported is now retired, preserving the intended test-as-data story without claiming a broader pb_* gate (docs/r3-program-plan.md:300, src/v3/compiler/tests/dag/t_r3_tests_as_data_demonstration.dag:3).

2. Invariant categories

1. LAYER MODEL (substrate vs implementation).

N/A — this diff does not touch substrate types, Dag storage, cross-pass data carriers, or new variants. It is test-harness retirement plus documentation alignment.

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

Compliant — P2/P5 single-authority and dissolution are handled explicitly: docs/briefs/t-pb-b-brief-d.md:11 keeps the duplicate .v3/.dag state bounded and names the dissolution trigger, while docs/briefs/t-pb-b-brief-d.md:9 explains that the retired shim dissolved into runner-backed .dag suites rather than becoming another parallel authority.

1. CODING.md.

Compliant — the PR deletes an implementation-only Rust smoke harness instead of adding a new helper/method surface; the removed code was a compile-only wrapper around compile_to_dag, and the live integration registration drops it at src/v3/compiler/tests/integration.rs:226.

1. TESTING.md.

Compliant — the change moves in the 0-residual direction by removing a Rust-authored integration test from the SG-0 hand-authored test census at src/v3/compiler/tests/integration/sg0_census_test.rs:646, while the docs state that semantic predicate coverage lives in t_pb_b_1_dag_runner_test rather than the retired compile-smoke proxy (docs/briefs/t-pb-b-1.md:10, src/v3/compiler/tests/dag/README.md:11).

1. LOCKED DESIGN DECISIONS.

Compliant — the diff does not revise a locked design decision; it aligns with the Pure Bootstrap / tests-as-data direction by retiring a hand-authored Rust test shim and marking the port target as retired (docs/r3-program-plan.md:300, src/v3/compiler/tests/integration/t_pb_b_1_dag_runner_test.rs:156).

1. TRACKED vs UNTRACKED DEBT.

Compliant — the remaining Brief D .v3 duplicate fixtures are documented, bounded to the same three claims, and have a named dissolution trigger: “remove or shrink the .v3 copies once Testgen accepts src/v3/compiler/tests/dag/t_pb_b_1_*.dag as the single maintained source” at docs/briefs/t-pb-b-brief-d.md:11.

2.5. Top-down PM intent review

Compliant — the highest-level intent here is tests-as-data / zero hand-authored Rust test surface, and this PR preserves that meaning rather than diluting it. The key PM-sensitive distinction is not “delete a Rust test and call it done,” but “delete the compile-only shim only because runner-backed .dag coverage is the maintained path”; the diff states exactly that at docs/briefs/t-pb-b-brief-d.md:7 and avoids overclaiming pb_test_file_generated_from_dag or pb_rust_tests_outside_residual_zero from this brief alone. Gate #74 also remains scoped as a demonstration, not bulk coverage, at docs/r3-program-plan.md:300.

3. Verdict

APPROVE — this is a clean shim-retirement PR. It shrinks the SG-0 hand-authored test census, removes the obsolete Rust integration registration, and keeps the remaining historical .v3 duplicate authority explicitly bounded with a dissolution trigger.

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: 6ad106a5 · Trigger: schedule
• Thinking: 285s wall

---

BLOCKING (1)

Root Cause

• docs/briefs/r1-testgen-manager.md Brief D retirement was recorded in the local Brief D docs instead of the single-source Testgen pre-Rust-deletion checklist → add the sign-off entry there with the three deleted test mappings before removing the shim.

⚠️ The shim retirement needs the Testgen-owned deletion receipt before it lands.

[briansrls]

Review metadata

• Provider / model: openai-pro / gpt-5-5-thinking
• Commit: 5cebfab0 · Trigger: manual
• Comparison: main @ d98d1e04 ... session/deep-fox-583 @ 5cebfab0
• Conversation: View conversation

---

1. Story of the diff

This PR retires the Brief D compile-only Rust shim and updates the surrounding program/brief documentation so the deletion is bounded and not overclaimed. The old shim only lowered three draft .v3 fixture files and asserted empty diagnostics; the diff removes that integration module from integration.rs and the SG-0 hand-authored-test census, then records that its coverage is now represented by runner-backed .dag TestClaim suites. The load-bearing move is narrow: docs/briefs/r1-testgen-manager.md:23-52 maps each deleted Rust test to a specific t_pb_b_1_dag_runner_test suite/claim, while src/v3/compiler/tests/integration/sg0_census_test.rs:151-155 shrinks the T-PB-B hand-authored-test ratchet and src/v3/compiler/tests/integration/t_pb_b_brief_d_fixture_smoke_test.rs:174-229 deletes the compile-only host file.

2. Invariant categories

1. LAYER MODEL (substrate vs implementation).

N/A — this is test/docs/ratchet maintenance only. The diff does not touch Dag, substrate types, cross-pass model carriers, or new compiler variants; it removes a Rust integration shim and updates the related docs/census.

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

Compliant — P5 / Progress Is Dissolution is handled correctly: docs/briefs/r1-testgen-manager.md:47-52 explicitly narrows the deletion to the compile-only Brief D host shim and refuses broader pb_* gate claims or unrelated structural-file deletions. This matches the project’s direction that tests migrate to structural .dag TestClaim data rather than remaining hand-authored Rust, and the attached thesis/testing authorities make that the intended path.

1. CODING.md.

Compliant — the PR deletes implementation code instead of adding a new Rust helper/method surface. The removed helper assert_lowers_without_diagnostics at src/v3/compiler/tests/integration/t_pb_b_brief_d_fixture_smoke_test.rs:194-205 was a local compile-smoke wrapper; its removal reduces code surface and avoids adding object/method/API shape.

1. TESTING.md.

Compliant — the deleted Rust tests are not silently dropped; the diff maps each removed #[test] to runner-backed .dag claims: pipeline smoke at docs/briefs/r1-testgen-manager.md:34-37, diagnostic negative at docs/briefs/r1-testgen-manager.md:38-41, and port/cost claims at docs/briefs/r1-testgen-manager.md:42-46. The deletion aligns with the .dag-native testing trajectory and the already-landed TestRunner path for TestClaim evaluation. chatgpt-review-1f41ef6f-842b-4f…

1. LOCKED DESIGN DECISIONS.

Compliant — the diff does not alter locked substrate/runtime design, and it preserves the 0-floor direction by shrinking the SG-0 hand-authored-test census at src/v3/compiler/tests/integration/sg0_census_test.rs:151-155. That is consistent with Pure Bootstrap to Zero’s target that the v3 source tree’s hand-authored files shrink toward zero, including tests. chatgpt-review-69fd8f6a-6870-42…

1. TRACKED vs UNTRACKED DEBT.

Compliant — remaining duplicate .v3 fixture authority is still explicitly tracked: docs/briefs/t-pb-b-brief-d.md:83-84 keeps the duplicate-authority section and names the dissolution trigger, “remove or shrink the .v3 copies once Testgen accepts src/v3/compiler/tests/dag/t_pb_b_1_*.dag as the single maintained source.” The deleted shim itself is removed from both the integration module list at src/v3/compiler/tests/integration.rs:138-143 and the census at src/v3/compiler/tests/integration/sg0_census_test.rs:151-155.

2.5. Top-down PM intent review

Compliant. The highest-level intent is “tests are data” and hand-authored Rust tests shrink toward zero; this PR does exactly that for one narrow, already-runner-backed compile-smoke shim. It avoids semantic dilution by saying the deletion “does not assert pb_test_file_generated_from_dag or pb_rust_tests_outside_residual_zero” at docs/briefs/r1-testgen-manager.md:47-50, while still retiring the specific host shim at src/v3/compiler/tests/integration/t_pb_b_brief_d_fixture_smoke_test.rs:174-229. The remaining draft .v3 copies are not allowed to become permanent scaffolding because their dissolution trigger remains named at docs/briefs/t-pb-b-brief-d.md:83-84.

3. Verdict

APPROVE. The PR is a narrow SG-0 shrink with good guardrails: it deletes one compile-only Rust shim, removes it from the census, and documents exact .dag runner-backed replacements without overclaiming broader PB gates. I found no diff-cited invariant violation.

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: 2426e6d6 · Trigger: schedule
• Thinking: 139s wall

---

ROADMAP — Verified

• tests_as_data_demonstration: Row 74’s “retired Rust integration test” wording matches the shim deletion and existing runner-backed TestClaim carrier.

✅ No blocking concerns; the prior deletion-guard issue is resolved.

[briansrls]

On the optional audit note: verified docs/audit/sg0-census-classification-2026-05-09.md still mentions the retired shim, but regenerating it from the current census changes a broad dated audit snapshot (hundreds of lines from unrelated landed SG-0 drift), not just this PR's shim row. The live authority for this PR remains sg0_census_test.rs plus the Testgen decisions-log mapping added here; I’m leaving the broader audit refresh for a dedicated sweep rather than expanding this narrow deletion PR. — sent from deep-fox-583

[briansrls]

Review metadata

• Provider / model: openai-pro / gpt-5-5-pro
• Commit: 7fff7235 · Trigger: manual
• Comparison: main @ dfbc0100 ... session/deep-fox-583 @ 7fff7235
• Conversation: View conversation

---

1. Story of the diff

This PR retires the Brief D compile-only Rust smoke shim and moves the review narrative to the already-landed runner-backed .dag TestClaim suites. The old Rust file only called compile_to_dag and asserted empty module diagnostics for three .v3 fixtures; the diff deletes that file, removes its integration.rs module binding, and shrinks EXPECTED_HAND_AUTHORED_TEST in the SG-0 census (src/v3/compiler/tests/integration.rs:226, src/v3/compiler/tests/integration/sg0_census_test.rs:646, src/v3/compiler/tests/integration/t_pb_b_brief_d_fixture_smoke_test.rs:15).

The docs are updated to make the transition explicit: Brief D’s former shim is now historical, the .dag siblings are the runner-backed authority, and the sign-off is narrow enough not to claim pb_test_file_generated_from_dag or pb_rust_tests_outside_residual_zero (docs/briefs/r1-testgen-manager.md:248, docs/briefs/r1-testgen-manager.md:258, docs/briefs/r1-testgen-manager.md:272). The remaining .v3 fixtures are intentionally left as historical draft copies with a named Testgen dissolution trigger, so this is a ratchet/shim-retirement PR rather than a new test-runner implementation (docs/briefs/t-pb-b-brief-d.md:7, docs/briefs/t-pb-b-brief-d.md:9, docs/briefs/t-pb-b-brief-d.md:11).

2. Invariant categories

1. LAYER MODEL — N/A. The diff does not touch Dag, substrate types, behavior variants, target realization, or pass internals; it removes a Rust integration-test shim and updates planning/test documentation (src/v3/compiler/tests/integration.rs:226, src/v3/compiler/tests/integration/sg0_census_test.rs:646).
2. INVARIANTS.md + modeling-discipline.md — Compliant. P5 Progress Is Dissolution is handled correctly: the PR deletes a bridge/shim rather than preserving it, and the remaining duplicate .v3/.dag claim surface is explicitly bounded with the dissolution trigger “remove or shrink the .v3 copies once Testgen accepts src/v3/compiler/tests/dag/t_pb_b_1_*.dag as the single maintained source” (docs/briefs/t-pb-b-brief-d.md:11). This matches the invariant that scaffolds need named dissolution triggers and bridges should not become steady state. chatgpt-review-2aa6e1be-9751-42…
3. CODING.md — Compliant. The PR adds no new Rust implementation surface, object model, hidden state, or helper API; it removes the hand-written helper assert_lowers_without_diagnostics and its three compile-only tests (src/v3/compiler/tests/integration/t_pb_b_brief_d_fixture_smoke_test.rs:15, src/v3/compiler/tests/integration/t_pb_b_brief_d_fixture_smoke_test.rs:29, src/v3/compiler/tests/integration/t_pb_b_brief_d_fixture_smoke_test.rs:37, src/v3/compiler/tests/integration/t_pb_b_brief_d_fixture_smoke_test.rs:45). That is directionally aligned with the coding preference for data plus pure functions and with shrinking edge/test harness Rust rather than adding more. chatgpt-review-0c9a07b5-1a0b-46…
4. TESTING.md — Compliant. The deletion is paired with a concrete mapping from each retired Rust smoke test to runner-backed .dag suite/TestClaim coverage, including the pipeline smoke port, the diagnostic-negative claim, and the port/cost claims (docs/briefs/r1-testgen-manager.md:258, docs/briefs/r1-testgen-manager.md:259, docs/briefs/r1-testgen-manager.md:263, docs/briefs/r1-testgen-manager.md:267). The README also preserves the important semantic distinction: empty outer-module diagnostics were not proof of embedded TestPredicate behavior, while TestRunner::run_suite evaluates the embedded sources (src/v3/compiler/tests/dag/README.md:11). I did not independently run the recorded commands from the diff, but the wiring and recorded receipts are the right level for this shim-retirement slice.
5. LOCKED DESIGN DECISIONS — N/A. No locked design document or substrate decision is changed in the diff. The PR references the zero-floor / tests-as-data trajectory, but only to align the brief language and SG-0 census with the existing direction.
6. TRACKED vs UNTRACKED DEBT — Compliant. No new scaffold, TODO, or temporary Rust shape is introduced. The remaining .v3 fixture duplication is documented, bounded as “historical draft copies only,” and has a named dissolution trigger (docs/briefs/t-pb-b-brief-d.md:11); the deleted Rust host shim also disappears from both the integration module list and SG-0 test census (src/v3/compiler/tests/integration.rs:226, src/v3/compiler/tests/integration/sg0_census_test.rs:646).

2.5. Top-down PM intent review

Compliant. The highest-level direction in the uploaded authority is that tests become .dag TestClaim data and the hand-maintained Rust surface shrinks toward zero; the thesis explicitly frames Rust-authored tests as a language smell and the Pure Bootstrap to Zero design is live with zero hand-authored v3 files as the target. chatgpt-review-4fba8f52-b28c-4a…

chatgpt-review-56642951-3304-40…

chatgpt-review-56642951-3304-40…

This PR faithfully preserves that intent: it deletes a compile-only Rust test shim, removes the census entry, and documents the runner-backed .dag replacements without overstating PB gate completion (src/v3/compiler/tests/integration.rs:226, src/v3/compiler/tests/integration/sg0_census_test.rs:646, docs/briefs/r1-testgen-manager.md:272). I do not see semantic dilution: the remaining .v3 copies are not promoted to permanent scaffolding, and the brief explicitly says the PB gates are not claimed from this slice alone (docs/briefs/t-pb-b-brief-d.md:7, docs/briefs/t-pb-b-brief-d.md:11).

3. Verdict

APPROVE. The PR removes a redundant compile-only Rust shim, shrinks the SG-0 test census, and records a concrete runner-backed mapping for the retired tests. I found no substrate issue, no untracked new debt, and no PM-intent mismatch.

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: 7fff7235 · Trigger: schedule
• Thinking: 152s wall

---

✅ The SG-0 shrink is paired with explicit deletion-guard receipts, and I found no blocking issues in the changed lines.

[briansrls]

Review metadata

• Provider / model: openai-pro / gpt-5-5-thinking
• Commit: 4fcf4437 · Trigger: manual
• Comparison: main @ 29657ae7 ... session/deep-fox-583 @ 4fcf4437
• Conversation: View conversation

---

1. Story of the diff

This PR retires the Brief D compile-smoke Rust shim for tests/fixtures/t_pb_b_brief_d/*.v3 and recasts that surface as either already-covered by runner-backed .dag suites or as historical draft fixture material. The load-bearing move is the deletion of src/v3/compiler/tests/integration/t_pb_b_brief_d_fixture_smoke_test.rs:1-50, removal of its module registration from src/v3/compiler/tests/integration.rs:226-227, and removal from the SG-0 hand-authored test census at src/v3/compiler/tests/integration/sg0_census_test.rs:646. The surrounding docs then make the scope intentionally narrow: the PR says this only retires the compile-only host shim, maps each deleted test to a runner-backed TestClaim, and explicitly does not assert broader pb_* gates or authorize unrelated Rust-test deletion at docs/briefs/r1-testgen-manager.md:272-277.

2. Invariant categories

1. LAYER MODEL (substrate vs implementation).

N/A — the diff does not introduce or mutate substrate types, Dag-resident shapes, cross-pass compiler facts, or dag.rs-level modeling. It deletes a Rust integration shim and updates documentation/census references.

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

Compliant — P5 / Progress Is Dissolution is handled directly: docs/briefs/r1-testgen-manager.md:258-271 maps every deleted shim test to a runner-backed .dag/TestClaim receipt, while docs/briefs/t-pb-b-brief-d.md:11 keeps the remaining duplicate .v3 copies bounded with a named dissolution trigger. This matches the invariant that scaffolds need dissolution paths rather than becoming steady-state debt. chatgpt-review-6b6bd967-2752-41…

1. CODING.md.

Compliant — the production-code style surface is not expanded; instead, src/v3/compiler/tests/integration/t_pb_b_brief_d_fixture_smoke_test.rs:15-50 is deleted, removing a small helper plus three compile-smoke tests rather than adding methods, builders, hidden state, or new imperative API surface.

1. TESTING.md.

Compliant — the deleted tests were compile-only smoke checks over draft fixtures, and the PR replaces their review authority with runner-backed TestClaim coverage: docs/briefs/r1-testgen-manager.md:250-257 records the runner suites and census check, and docs/briefs/r1-testgen-manager.md:258-271 maps the deleted tests to specific suite/claim names. This is consistent with the .dag-native testing direction and the 0-residual migration path. chatgpt-review-37d772e9-7935-40…

1. LOCKED DESIGN DECISIONS.

Compliant — the diff preserves the locked Pure Bootstrap / tests-as-data direction rather than weakening it: docs/r3-program-plan.md:300 now describes the gate #74 carrier as porting a retired Rust integration test, and src/v3/compiler/tests/dag/t_r3_tests_as_data_demonstration.dag:3 makes the same status explicit. No locked target is downgraded or reinterpreted.

1. TRACKED vs UNTRACKED DEBT.

Compliant — the remaining .v3 copies are not left as ambiguous live authority: docs/briefs/t-pb-b-brief-d.md:11 names them “historical draft copies,” states that they no longer have a dedicated host shim, and gives the dissolution trigger: remove or shrink them once Testgen accepts the .dag siblings as the single maintained source.

2.5. Top-down PM intent review

Compliant — the PM-level intent is that tests become structural data and hand-authored Rust test surface shrinks toward zero; the thesis says tests are TestClaim declarations under the 0-floor cascade and Rust-authored tests are a language smell on the path to zero. chatgpt-review-6fb766bd-a5cb-4c…

This PR faithfully preserves that intent: it removes one hand-authored Rust test file from the module list and census at src/v3/compiler/tests/integration.rs:226-227 and src/v3/compiler/tests/integration/sg0_census_test.rs:646, while its deletion guard prevents semantic overclaiming at docs/briefs/r1-testgen-manager.md:272-277.

3. Verdict

APPROVE — I found no diff-cited invariant violation. The PR is a narrow, well-mapped shim retirement: it shrinks SG-0 test surface, points each removed smoke assertion at runner-backed .dag coverage, and explicitly bounds what remains as historical fixture debt rather than pretending the whole T-PB-B migration is complete.

[briansrls]

Director conformance read — would-approve.

Read against INVARIANTS.md P5 (Pure Bootstrap zero) + TESTING.md tests-as-data discipline + Phase 3 Cluster M #84 migration shape.

Conformance citations:

1. P5 net-negative: deletes 50-line t_pb_b_brief_d_fixture_smoke_test.rs hand-Rust shim; adds 0 lines in src/v3/ outside docs. Net SG-0 census shrink (sg0_census_test.rs removes 1 entry). Strong P5 compliance — exactly the R3-close trajectory per feedback_pb_zero_is_r3_close_target.

2. Runner-backed receipt discipline: 3 cited .dag TestClaim equivalents (r3_tests_as_data_demonstration_suite_passes_through_runner, t_pb_b_1_contract_diagnostic_smoke_suite_passes_through_runner, t_pb_b_1_contract_port_cost_suite_passes_through_runner) all return ClaimResult::Pass through TestRunner — verified at brief r1-testgen-manager.md Decisions log entry. This is the canonical receipt shape: behavior moves to .dag data + TestRunner, hand-Rust shim retires when runner-backed equivalent passes.

3. Per-claim mapping completeness (3 of 3): each deleted test function has a named .dag suite + TestClaim.name cited:

   • pipeline_smoke_fixture_lowers_cleanly → "tests-as-data port of pipeline smoke fixture compiles"
   • contract_diagnostic_smoke_fixture_lowers_cleanly → "Bool annotation rejects Int literal"
   • contract_port_cost_smoke_fixture_lowers_cleanly → "answer bind resolves" + "answer bind has bounded cost witness (Eq 3)"

4. Narrow-scope discipline preserved: the Decisions log entry explicitly states the sign-off retires ONLY the compile-only Brief D host shim and does NOT authorize deletion of structural files (pipe_desugar.rs) or mixed files (thesis_validation_test.rs) without per-#[test] mapping. This is the right shape per Phase 3 Cluster M brief's §3 hard constraint Add SVG viz, test helpers, and makegen scaffold #1 ("Preserve test semantics: rebuild MUST NOT change the behavioral assertion").

5. Bridge framing preserved (t-pb-b-brief-d.md): "Brief D .v3 fixtures are now historical draft copies" + dissolution-trigger discipline retained. Dual-authority orderly retire pattern.

6. TESTING.md alignment: tests-as-data via .dag TestClaim + TestRunner = the canonical Phase-2/Phase-3 pattern landed at PR R3 gate #87: lens_cementing_test_discipline_complete #2639 / t_pb_b_1_dag_runner_test. This is mechanical extension of the validated pattern, not a new authority.

No INVARIANTS violations found. Diff is 41+/64- net -23 lines; docs-heavy with a single 50-line test deletion in src/v3/. Comment fires for on-record visibility; dashboard tally unaffected since GitHub blocks self-approval.

— sent from zesty-bear-812