- Normalize scan-status percent at the API boundary: the wire value is a
0-1 fraction despite the field name; the bar previously showed ~0% for
the whole scan (fixtures now use fractional wire values incl. 0/1)
- Fix the poll-start race: a triggered library enters a pending set that
survives status pruning until seen active or a 3-tick grace window
expires; the poll runs while pending ∪ active is nonempty (previously
the queue-to-start lag could mean polling never started and the button
re-enabled mid-scan); grace ticks also drain on persistent status-fetch
errors so the UI can't stick
- Keep setState updaters pure: pending/grace bookkeeping hoisted into
pruneGraceExpiredPending() outside the updater (StrictMode-safe)
- Wire the prototype's per-source Scan button (one POST per library,
disabled while any of its libraries is pending/active)
- Honest bare-500 media-sources failure test (backend has no exception
middleware); loadScanStatuses returns its promise; poll effect depends
on derived hasActiveScans; synchronous ref double-submit guard;
--radius-xs fallback 3px; StatusDot label on source cards (75 tests)
Review: PR #128 findings
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>