workflow-fix: flaky settle-sleep equality assert in inline-l

Overview / Motivation

Auto-filed by the /daily 2026-08-01 Step C parked-candidate sweep from a workflow-fix candidate parked on task #1977 (emitting agent: implementer round 1, recursion-guarded; formal candidate block, fingerprint ffff2ff64004).

Goal

Harden test_write_cert_malformed_rehash_delay_falls_back_and_still_toctous (and its sibling assertion) against interpreter-internal sleeps captured by the process-global time.sleep monkeypatch, so unrelated machine load cannot flip the Step 9c gate red on tests/test_inline_lint_gate.py.

Workflow gap

• Bug observed: test_write_cert_malformed_rehash_delay_falls_back_and_still_toctous fails intermittently under load (emitter: 2 of 3 single-test runs in the issue-1977 worktree; passes at the main checkout) because monkeypatch.setattr(ilg.time, "sleep", ...) patches the GLOBAL time module and captures interpreter-internal backoff sleeps (a 0.001*2^k series, 105+ entries) alongside write_cert's single settle sleep, breaking the exact slept == [2.0] assertion.
• Why it is a workflow gap: the test pins the #1857 cert-retry workflow invariant on the Step 9c gate surface with an over-strict equality on a process-globally patched function, so unrelated machine load can flip the gate red on this file.
• Confidence (emitter): high
• verified-at-filing: grep -n 'slept' tests/test_inline_lint_gate.py around the cited lines → assert slept == [2.0], slept present at ~L594 and assert len(slept) == 1, slept at ~L573, both via monkeypatch.setattr(ilg.time, "sleep", ...) (2026-08-02 UTC). Landed-fix check: git log --oneline --since='7 days ago' -- tests/test_inline_lint_gate.py → 2 commits (#1889, #1857), neither filtering the sleeps. The intermittency itself is the emitter's measurement (unverified hypothesis — verify at plan time: 2-of-3 under-load failure reproduces; the mechanism — global patch capturing interpreter-internal 0.001*2^k backoff sleeps — is the emitter's traced diagnosis, re-verify by inspecting the captured slept list on a failing run).

Proposed change (candidate diff sketch — refine in planning)

- assert slept == [2.0], slept
+ settle = [s for s in slept if s >= 1.0]
+ assert settle == [2.0], slept

(same hardening for the sibling len(slept) == 1 assertion at ~L573.)

Scope / surfaces

• Primary target: tests/test_inline_lint_gate.py

Constraints / invariants

• Workflow-surface only. Ruff on touched files passes; the hardened asserts still pin the #1857 cert-retry invariant (exactly one 2.0s settle sleep from write_cert).
• Recursion guard applies (workflow_fix_target Provenance line below).

Provenance

• workflow_fix_target: tests/test_inline_lint_gate.py
• fingerprint: 81fc8443f020 (tag-authoritative; supersedes body-carried fingerprint: ffff2ff64004)
• origin: parked candidate on task #1977, ts 2026-08-01T08:06:03Z, routed by /daily 2026-08-01 Step C.

target_file: tests/test_inline_lint_gate.py bug_observed: test_write_cert_malformed_rehash_delay_falls_back_and_still_toctous fails intermittently under load (2 of 3 single-test runs in the issue-1977 worktree; passes at the main checkout) because monkeypatch.setattr(ilg.time, "sleep", ...) patches the GLOBAL time module and captures interpreter-internal backoff sleeps (a 0.001*2^k series, 105+ entries) alongside write_cert's single settle sleep, breaking the exact slept == [2.0] assertion at tests/test_inline_lint_gate.py:594. why_workflow_gap: the test pins the #1857 cert-retry workflow invariant on the Step 9c gate surface with an over-strict equality on a process-globally patched function, so unrelated machine load can flip the gate red on this file. proposed_change: assert the settle sleep by filtered membership (settle-scale sleeps only) instead of exact list equality, and apply the same hardening to the sibling len(slept)==1 assertion at tests/test_inline_lint_gate.py:573. confidence: high related_task: #1977