[leehack]

Fixes #1892

Summary

• Root cause: the supervisor set done_at = now, where now was read at the top of the loop before process_table(), so the post-done window started before the reporter_done event was logged. Recovery could fire up to one loop iteration (slow ps under load) less than post_done after the logged event: 0.598 vs 0.6 in CI. This was a real supervisor bug, not a test measurement error (both sides already use the monotonic clock).
• Fix: take done_at from time.monotonic() right after recording reporter_done. The test compares at the events' millisecond rounding precision, so float rounding of the rounded timestamps cannot fail it.

Production-readiness scope

• User-facing scope: None; CI tooling only.
• Supported platforms/paths: Linux CI Chrome supervisor.
• Unsupported or intentionally unavailable paths: N/A
• Out of scope / follow-ups: None

Completeness checklist

• Declared scope is fully implemented.
• Unsupported platform/option combinations: N/A (no behavior surface change).
• Public API docs, README/website docs, examples, support matrices, changelog: N/A, not user-visible.
• Regression coverage: existing test test_recovery_happens_no_earlier_than_post_done_seconds_after_done; mutant below.
• Validation coverage registry: N/A.
• Security/privacy review: no secrets involved.
• Follow-ups: None.

PR type guidance

• Bugfix PR: root cause above; regression test is the existing one, which fails when post_done is halved in the supervisor (0.352 not >= 0.6).

Test Plan

• dart run tool/prepare_workspace.dart
• dart format --output=none --set-exit-if-changed .
• dart analyze
• dart test -p vm -j 1 --exclude-tags local-only (CI at 34f8911d: Test Linux VM 1–3 green)
• dart test -p chrome --exclude-tags local-only N/A, CI tooling only
• coverage-lib N/A, no lib/ change
• Other: test_supervise_chrome_test.py (Linux-only) run in python:3.12-slim docker, repeated 30x with --cpus 1; macOS skips it.

Matrix Evidence

Matrix row	Scope covered	Platform / model / backend	Result	Evidence / notes
high-risk tier (regressionPolicy surface)	CI supervisor timing only	Linux docker, --cpus 1 under stress-ng	Pass	Independent audit PASS at 34f8911d: head 80/80 pass under load; base 3/80 fail (reproduces the flake); old done_at = now 9/80 fail; halved post-done 80/80 fail. Evaluator exit 2.

Review Notes

• Independent review status: independent audit PASS at 34f8911d, no blocking findings; follow-up (deterministic injected-delay regression test) logged on 0.19.x audit follow-ups (index) #1861
• CI status / head SHA: green at 34f8911daaf2e8ff1ed29a82962d74cbd06b27c9 (33 pass, 8 skipped)

High-risk regression review

• Classification: high-risk (surface: regressionPolicy)
• Implementation task: this PR
• Independent blocking QA task: fresh auditor session, PASS, 0 PR-caused P1, 0 unresolved threads
• Exact head / current base: 34f8911d / main 77b01da1f
• Production-branch deletion, bypass, or miswire proof: N/A
• Affected-family real-model/artifact evidence: N/A, no model-dependent behavior
• Explicit unavailable-family or other N/A evidence: N/A
• Known PR-caused P1 regressions: 0
• Unresolved review threads: 0

[github-actions]

Chat app preview removed for leehack/llamadart-chat-pr-1893.

[leehack]

Landed on main via batch integration PR #1895 (rebase merge, main ba4e0e1b; tree identical to the audited integration head 5ffd1224).