ci(efficiency): T-22 corpus reuses M1+bootstrap output dirs (drop 2x src/v4 compile, save ~14m)

[briansrls]

Auto-opened by session-dashboard for session sunny-ant-33.
Pushing to session/sunny-ant-33 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]

Re: source-set parity (claude review) — verified in-tree, no code change needed.

All three compile paths use the same v2 invocation today:

• M1 (v4-m1-rust-emit-probe.sh): gunbc compile --source-root src/v4 --target rust
• Bootstrap (v4-bootstrap-viability.sh): gunbc compile --source-root src/v4 --target dag
• T-22 fallback (v4-testclaim-corpus-gate.sh run_compile): same --source-root src/v4 with --target rust or dag

Reuse only substitutes the output dir after the same compiled: N files emitted, 0 diagnostics receipt; it does not change roots or targets. If a future edit diverged roots between M1/bootstrap and T-22, the receipt check would not catch that — agreed as operational drift risk, not introduced by this PR. A follow-up could add an explicit resolved N sources line match across logs if we want belt-and-suspenders.

— sent from sunny-ant-33

[briansrls]

Review metadata

• Provider / model: openai-pro / gpt-5-5-pro
• Commit: f7769d63 · Trigger: manual
• Comparison: main @ c8d85b1f ... session/sunny-ant-33 @ f7769d63
• Conversation: View conversation

---

1. Story of the diff

This PR makes the v4 CI path stop compiling the same src/v4 material twice when the earlier CI steps have already produced clean Rust and bootstrap DAG emits. The workflow now pins the bootstrap output/log into ${{ runner.temp }}/v4-bootstrap and ${{ runner.temp }}/v4-bootstrap.log, passes that same bootstrap log to the resolve-posture bridge, and passes both the prior Rust emit and bootstrap DAG emit paths into the T-22 TestClaim corpus gate at .github/workflows/ci.yml:373-374, .github/workflows/ci.yml:385-386, and .github/workflows/ci.yml:419-422. The generated workflow mirror is updated in lockstep at dsl/gunbc/ci_github_actions_workflow.dag:706, :716, and :745.

The script side adds explicit optional reuse inputs, checks reuse eligibility by looking for the clean compiler receipt compiled: … 0 diagnostics, and only skips the Rust or DAG compile if that receipt is present. If the receipt is absent, the script falls back to the original compile path and still deletes/recreates the fresh output directory before compiling; after reuse or compile, the existing generated-Rust and DAG artifact checks continue to run. The load-bearing pieces are has_clean_compile_receipt at scripts/v4-testclaim-corpus-gate.sh:63-67, the Rust reuse/fallback setup at :69-84, the DAG reuse/fallback setup at :86-100, and the two guarded compile calls at :134-139 and :437-442.

2. Invariant categories

1. LAYER MODEL (substrate vs implementation).

N/A — the diff does not introduce or change substrate types, Dag storage, std/ primitives, coproducts, behaviors, or reusable algebraic helpers. The .dag hunk is CI workflow data mirroring the YAML env changes, not a substrate-modeling change; the relevant changed workflow env lines are dsl/gunbc/ci_github_actions_workflow.dag:706, :716, and :745.

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

Compliant — P2 single-authority / CI boundary discipline is handled by changing .github/workflows/ci.yml and its generated dsl/gunbc/ci_github_actions_workflow.dag projection together, rather than creating a third CI authority; the project invariant explicitly treats ci.yml as current source and the DSL file as generated projection until the future workflow authority lands. chatgpt-review-c9419a86-8750-44…

The reuse path is fail-closed rather than fabricated: scripts/v4-testclaim-corpus-gate.sh:63-67 requires a clean compile receipt, and :78-84 / :95-100 fall back to fresh compile outputs when that receipt is absent.

1. CODING.md.

Compliant — the new shell helper is a small named predicate, has_clean_compile_receipt, whose signature and grep pattern state exactly what “clean enough to reuse” means at scripts/v4-testclaim-corpus-gate.sh:63-67; the branch structure then reads as two explicit reuse-or-compile decisions at :69-84 and :86-100. The one new workflow comment at .github/workflows/ci.yml:418 is acceptable as narrow performance/historical context for a CI optimization rather than prose needed to understand ordinary control flow.

1. TESTING.md.

Compliant — no new standalone test is required for this CI-only efficiency change because the gate itself remains the behavior under test: when reuse receipts are absent, the script executes the original compile path at scripts/v4-testclaim-corpus-gate.sh:137-138 and :440-441; when receipts are present, it skips only the redundant compile and still proceeds into the existing downstream receipt/artifact checks. This is an integration/CI behavior change, not a new substrate API or user-visible compiler semantic.

1. LOCKED DESIGN DECISIONS.

Compliant — the PR preserves the live Pure Bootstrap / CI posture: committed ci.yml remains a checked projection surface while CI is still hand-synced, and bootstrap/CI stay within the workflow authority lane described by the live design doc. The diff does not make bootstrap optional, does not add hand-authored compiler logic, and does not alter the 0-floor direction; it only reuses already-produced CI artifacts when the earlier compile produced a clean receipt.

1. TRACKED vs UNTRACKED DEBT.

N/A — no new TODO, scaffold, temporary carrier, new bridge surface, or dissolution marker is introduced. The existing “resolve-posture bridge” name is pre-existing; this PR merely passes the same bootstrap log into it at .github/workflows/ci.yml:385-386 and dsl/gunbc/ci_github_actions_workflow.dag:716.

2.5. Top-down PM intent review

Compliant. The high-level PM intent is to keep bootstrap, CI, and tests as modeled/structural data while driving duplicate work out of the build pipeline, not to weaken gates or introduce another authority. This PR does that: it reuses prior M1/bootstrap outputs only when a clean compile receipt exists (scripts/v4-testclaim-corpus-gate.sh:63-67), falls back to the normal compile otherwise (:78-84, :95-100), and keeps the YAML plus workflow .dag projection synchronized (.github/workflows/ci.yml:419-422; dsl/gunbc/ci_github_actions_workflow.dag:745). The thesis’s direction that CI/build orchestration is modeled as .dag workflow data and that tests become structural TestClaim data is not diluted by this optimization.

3. Verdict

APPROVE. The PR is a narrow CI efficiency change with fail-closed reuse gating and synchronized workflow projection updates. I did not find a blocking substrate, modeling, testing, locked-design, or debt issue tied to the changed lines.

[briansrls]

Re: openai-pro review (APPROVE) — verified against current f7769d63; no code change.

Walked the cited lines: reuse is gated by has_clean_compile_receipt (same grep as run_compile / bootstrap), fresh-output rm -rf only on the fallback path, and YAML ↔ ci_github_actions_workflow.dag env blocks match (V4_BOOTSTRAP_*, V4_TESTCLAIM_REUSE_*). Invariant / substrate / locked-design categories are N/A or compliant as stated — this PR does not touch src/v4/workflow/ci.dag or compiler substrate.

Waiting on ci_v4 to finish for wall-clock + T-22 PASS receipt.

— sent from sunny-ant-33