[bug] attempts.ended_at is never written #66

Summary

attempts.ended_at is never written. Attempts that have finished, whether they failed or reached AwaitingReview, keep ended_at = NULL.

Version / OS

meshloop built from 09a823a (0.3.0). Windows 11 native.

Observed

In the store, after graphs issues-46-47-56 and issues-56 finished:

runtime-design.md §4 lists ended_at as part of the attempt record.

Impact

• PR fix: status honours --config, observed liveness for unowned workers, record skipped verification #63 ([bug] inspect reports live: Dead for detached-run workers that are running #47) stops checking a process once ended_at is set. Because ended_at is never set, a finished worker's PID is still checked, and a reused PID can read as Ambiguous. On POSIX, where image names aren't compared, it can read as Live.
• Attempt duration can't be worked out from the store, which [feature] inspect: expose per-attempt model, timing, outcome and evidence #57 needs.

Expected

The run loop sets ended_at when an attempt's harness exits, times out, or is cancelled. That is the same point where HarnessExited, HarnessCrashedOrTimeout or a cancel is recorded.

Acceptance criteria

• Fixture test: after a fixture run reaches AwaitingReview, its attempt has ended_at set, and it is not earlier than started_at.
• Test: a cancelled or timed-out attempt also gets ended_at.

🤖 Generated with Claude Code