[rmstdope]

Bead: cb-yf8h. Testing: R27's run waits out the 30-second end grace for four passes.

In tests/fleet-role-triggers.sh, run 2 (R27) no longer waits out the real 30 s end grace for its four clock passes before q. While a $FIXTURE/backdate marker exists, the stub claude dates its waiting ten minutes back right after end-pass. It writes the date beside the state file and moves it over, which is fleet-view-restart.sh's cb-n53p pattern. So the view ends each pass on its next tick. The marker is removed before run 3, which still measures R14's 30–50 s grace on undated passes.

Validation

• time bash tests/fleet-role-triggers.sh on a warm build, locally: 156 s on main, 118 s with this change. All assertions pass, including R14's 30–50 s window in run 3 and R27's starts in run 2.
• bash tests/gate: the Cargo, UI and bash legs are green. The one exception was fleet-keys.sh and fleet-type.sh, which timed out waiting for "the stub agent to start" while other sessions were building on this machine. Both pass when run alone, and this diff does not touch them.
• After the merge: the suite's time on a main CI run should be near ~100 s (it is 146 s on run 42803911818).

Deviation from the plan

tests/slow-suites: the figures are refreshed from this PR's CI run, once it has run (see the follow-up commit).

🤖 Generated with Claude Code

https://claude.ai/code/session_01WCDKeyim8VMAD4KoXKzjHn

[Copilot]

Note

Copilot was unable to complete this review at the selected tier due to model policy or plan restrictions.

🟡 Changes recommended

A moderate active issue remains in tests/fleet-role-triggers.sh where the new backdate block can corrupt the JSON state by moving an empty/partial tmp file if jq fails, risking flaky test runs until the mv is guarded on jq success with proper cleanup.

1 open issue

• Only replace state file after jq succeeds · New

What changed in this PR

Speeds up the tests/fleet-role-triggers.sh system suite by avoiding a real-time 30s end-of-pass grace wait in run 2 (R27) via conditional backdating of the stub agent’s waiting.since, while keeping run 3 as the real grace-period measurement.

Changes:

• Add a backdate marker-driven path that rewrites waiting.since ~10 minutes into the past after end-pass so passes end on the next tick.
• Create/remove the backdate marker around run 2 to limit the behavior to the intended segment.
• Update suite comments to document the new timing behavior and rationale.
• Address review notes: ensure the state rewrite cannot corrupt the JSON if jq fails (only mv on success and clean up temp), and tighten run 2’s upper-bound timing so the speedup is asserted rather than silently regressing.

File	Description
tests/​fleet-role-triggers.sh	Introduces a fixture-gated backdating mechanism for the stub agent’s waiting.since to shorten run 2 runtime while preserving run 3’s real end-grace timing assertions.

_{Review effort: Balanced}

[rmstdope]

Fixed in 8dd38a2: the mv now runs only when jq succeeds (the stub has no set -e). A failed rewrite leaves the state file whole, and run 2's 20 s end wait then fails, naming the pass.

[rmstdope]

Review (cold read, reviewed_head: 00d798f73a3df5aceea7fc403900ed86d3c503d4): reviewed before the merge under the producer's review practice, by an agent given the diff and the bead and not the producer's reasoning.

Verdict: nothing blocks the merge. The marker is scoped to run 2 (reset_run clears waited/* and the state files between runs), so run 3's R14 window is still measured on undated passes. Backdating since is the right lever (lifecycle.rs:271).

1. tests/fleet-role-triggers.sh: run 2's end wait still allows 70 s. If the backdating stopped working, the suite would just pay the half minute again and pass, so the speed-up could regress with no failure. A bound of about 10–15 s would make the suite hold its own speed-up.
2. tests/slow-suites still reads fleet-role-triggers.sh # 81s. The plan's increment 2 asked for a refresh, and the order will likely stay the same.

Answers (Cyclops, at the next head):

1. Changed. Run 2's end wait is now wait_until 20, with a comment that says why. Locally the suite is green at that bound, in 109 s.
2. Changed. The figures are refreshed from this PR's own CI run 42858875027, where fleet-role-triggers.sh took 114 s (146 s on main's run 42803911818). The order is unchanged at the top. fleet-verification-skill.sh (6 s on CI) drops off, and launch-preflight, install and agent-state come in. Both answers are the reviewer's own suggestions, applied as given, so no follow-up review is needed.