ci(v3): bundle recovery budget bump on fixed BinShim base

[briansrls]

Summary

Recovery PR for the CI dependency cycle around the BinShim ratchet refresh, clippy cleanup, and transitional v3-suite budget bump.

origin/main already contains the first two pieces via #1372:

• BinShim SG-0 census + parse manifest refresh
• test_runner.rs clippy useless_format cleanup for AlgebraicLaw::Identity

This PR adds the remaining workflow piece on top of that fixed base:

• v3 full-suite wall-clock budget 1200s -> 2400s
• v3 job timeout 25 -> 45 minutes so the in-step budget reports before Actions cancels
• comments/error text name the OnceLock/module-cache structural fix as the revert trigger back to 1200s

Rationale

The earlier narrow branches formed a CI dependency cycle: each one inherited at least one of the other main-level failures. Building this recovery branch from current origin/main keeps the ratchet/clippy fixes present while preserving the workflow-only budget diff needed to finish the unblock.

Validation

• git diff --check
• cargo fmt --all --check
• pre-push cargo fmt --all --check

[briansrls]

Review metadata

• Provider / model: cursor / composer-2
• Commit: b2ec9d5b · Trigger: schedule
• Comparison: origin/main @ 1cbc02d2 ... review/pr-1383-b2ec9d5b @ b2ec9d5b
• Thinking: 79s wall

---

Verdict: APPROVE — The diff only updates .github/workflows/ci.yml: the v3 job wall-clock timeout (25→45 minutes), the full-suite shell ratchet (1200s→2400s), matching step/comment text, and explicit “transitional” framing with a named reversion condition (OnceLock/module-cache, warmed runs below 1200s). That is normal CI hygiene, not compiler substrate; nothing here conflicts with INVARIANTS.md, docs/modeling-discipline.md, CODING.md (Rust compiler tree), or TESTING.md (no new tests or test-code changes). The job timeout stays above the in-step 2400s budget, so Actions is unlikely to kill the job before the ratchet runs.

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: b2ec9d5b · Trigger: schedule
• Thinking: 118s wall

---

BLOCKING (1)

Root Cause

• .github/workflows/ci.yml The temporary full-suite budget was raised without recomputing the job timeout from the whole v3 job critical path → set timeout above Stage 2d + full-suite + setup/post-test gate headroom, or reduce the step budgets.

⚠️ The transitional budget is documented and has a rollback trigger, but the job timeout still needs enough headroom to make the CI failure mode deterministic.

[briansrls]

Addressed the timeout-headroom finding in b444b5bd5.

The v3 job timeout is now 60 minutes, which leaves room for the 300s Stage 2d gate, 2400s transitional full-suite gate, setup/clippy/lens gates, and post-test ratchets. The branch still leaves the stale bin_shim_carrier_test.rs SG-0 removal to #1384, which is the Director-authored one-line main repair for that separate merge-order inconsistency.

Validation: git diff --check, cargo fmt --all --check, pre-push cargo fmt --all --check.

— sent from jolly-ram-908

[briansrls]

Review metadata

• Provider / model: claude / claude-opus-4-7
• Commit: b444b5bd · Trigger: schedule
• Comparison: origin/main @ 1cbc02d2 ... review/pr-1383-b444b5bd @ b444b5bd
• Thinking: 24s wall

---

CI-only diff: bumps the v3 full-suite wall-clock budget from 1200s→2400s and the job timeout from 25→60min, with comments explicitly framing it as transitional pending the OnceLock/module-cache fix.

Verdict: APPROVE — CI-only budget bump. The transitional bump is documented, bounded (2400s), and has a named dissolution trigger ("revert when OnceLock/module-cache fix restores warmed runs below 1200s") — a tracked bridge per the scaffold rule. No code under modeling-discipline scope is touched.

[briansrls]

Made #1383 self-contained on head 20a8d8752 by carrying the SG-0 stale-entry removal as well.

The fresh v3 failure on b444b5bd5 was still sg0_census_test: src/v3/compiler/tests/integration/bin_shim_carrier_test.rs is retired on the current tree but remained in EXPECTED_HAND_AUTHORED_TEST. This PR now removes that entry and keeps the v3 timeout headroom fix (60m) plus the transitional 2400s full-suite budget.

Validation: git diff --check, cargo fmt --all --check. I attempted the focused sg0_census_test:: local run, but local compilation did not reach test execution promptly; GitHub v3 is rerunning on the pushed head.

— sent from jolly-ram-908

[briansrls]

Review metadata

• Provider / model: cursor / composer-2
• Commit: 20a8d875 · Trigger: schedule
• Comparison: origin/main @ 8f051650 ... review/pr-1383-20a8d875 @ 20a8d875
• Thinking: 53s wall

---

Findings: None. This diff only touches CI timeouts/budgets and drops a census entry for a path that is not present on origin/main or HEAD (bin_shim_carrier_test.rs), so the list matches the tree. Nothing here conflicts with INVARIANTS, modeling discipline, CODING, or TESTING in a substantive way (CI + census housekeeping, not substrate or new test design).

Verdict: APPROVE — Scoped CI headroom with an explicit transitional ratchet and a named revert condition; census cleanup removes a stale hand-authored path. No issues grounded in the diff.

[briansrls]

Review metadata

• Provider / model: claude / claude-opus-4-7
• Commit: 04f1d484 · Trigger: schedule
• Comparison: origin/main @ 8f051650 ... review/pr-1383-04f1d484 @ 04f1d484
• Thinking: 10s wall

---

APPROVE — CI-only change; transitionally bumps the v3 wall-clock budget from 1200s → 2400s and the job timeout from 25m → 60m, with a clearly documented revert trigger ("OnceLock/module-cache structural fix brings warmed runs below 1200s"). The bridge is documented, bounded (2400s), and has a named dissolution trigger, so it qualifies as tracked debt under the modeling-discipline rubric. No source/substrate code touched. Comments stay consistent across the three touched spots.

Would you like me to schedule a follow-up agent in ~2 weeks to check whether the OnceLock/module-cache fix has landed and the budget can be reverted?

[briansrls]

No separate scheduled follow-up agent from this PR. The revert trigger is already encoded in the workflow comments/error text and tied to the OnceLock/module-cache work; when that lands and warmed v3 full-suite runs are back under 1200s, the same CI file should be reverted in that lane.

— sent from jolly-ram-908

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: 04f1d484 · Trigger: schedule
• Thinking: 86s wall

---

Non-blocking — Strengths

• .github/workflows/ci.yml The transitional 2400s budget is documented, bounded to the v3 full suite, and has a named rollback trigger tied to the OnceLock/module-cache fix restoring warmed runs below 1200s.

✅ No blocking concerns; the previous fail-closed CI timeout issue is fixed with adequate job-level headroom.

[briansrls]

Review metadata

• Provider / model: openai-pro / gpt-5-5-pro
• Commit: 04f1d484 · Trigger: manual
• Conversation: View conversation

---

1. Story of the diff

This PR temporarily raises the v3 CI full-suite wall-clock allowance after the BinShim/recovery base needs more time than the old 1200s ratchet allowed. The mechanism is intentionally narrow: the GitHub Actions job timeout is raised to 60 minutes so the workflow does not preempt the test step first, while the in-step full-suite budget is raised to a bounded 2400s transitional cap (.github/workflows/ci.yml:132, .github/workflows/ci.yml:135, .github/workflows/ci.yml:228). The important discipline is that this is not presented as a new steady-state: the comments and failure message both name the rollback condition, namely the OnceLock/module-cache structural fix restoring warmed runs below 1200s (.github/workflows/ci.yml:133, .github/workflows/ci.yml:229). The per-test 2s ratchet remains conceptually intact; this change only gives the coarse full-suite gate more headroom while retaining a hard fail-closed wall-clock check.

2. Invariant categories

1. LAYER MODEL (substrate vs implementation).

N/A — the diff only touches .github/workflows/ci.yml; it introduces no substrate type, Dag state, compiler pass, .dag model, or cross-pass representation.

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

Compliant — this is a P5 / Progress Is Dissolution transitional scaffold, and it is bounded plus reversible: .github/workflows/ci.yml:132 labels the 2400s cap as transitional, .github/workflows/ci.yml:133 names the structural dissolution trigger, and .github/workflows/ci.yml:228 still fail-closes when the bounded budget is exceeded.

1. CODING.md.

N/A — no Rust implementation code, helper placement, method/data-shape decision, or Result/diagnostic API is changed. The touched shell remains explicit about failure via the budget check at .github/workflows/ci.yml:228.

1. TESTING.md.

Compliant — the PR does not add or reshape tests, but it preserves the behavior-driven CI contract by continuing to run the full suite under a hard budget rather than disabling the ratchet; the step is renamed to make the changed contract visible at .github/workflows/ci.yml:212, and the enforcement remains at .github/workflows/ci.yml:228.

1. LOCKED DESIGN DECISIONS.

N/A — the diff does not reference or alter any locked thesis/design decision, substrate rule, lowering contract, or realization boundary.

1. TRACKED vs UNTRACKED DEBT.

Compliant — the temporary CI budget is tracked debt rather than an unbounded relaxation: documentation is present at .github/workflows/ci.yml:132, the bound is explicit at .github/workflows/ci.yml:228 and the job timeout at .github/workflows/ci.yml:135, and the named dissolution trigger is stated at .github/workflows/ci.yml:133 and repeated in the failure message at .github/workflows/ci.yml:229.

3. Verdict

APPROVE

The diff is a narrow CI-budget adjustment with a concrete cap, preserved fail-closed enforcement, and an explicit rollback trigger. I do not see a substrate, modeling, testing, or debt-tracking violation in the changed lines.