- Add replay regression test suite covering irregular/jittered timestamps, identical/zero dt, inverted timestamps, large gaps, pause/resume timing reconciliation, manual seek, speed changes, and large datasets (144k points).
- Harden advancePlaybackIndex against non-finite/NaN budget and timestamp delta anomalies.
- Reset lastTimeRef and cap frameDelta in WebTelemetryReplayModal to eliminate sudden jumps on speed switching or background tab resumes.
- Guard group-ride creation with in-flight isSubmitting state and disabled submit button in CreateGroupRideModal; guarantee unique collision-free ride IDs across concurrent submissions.
- Remove obsolete Vercel reference in website/.env.example in favor of Vultr VPS and Caddy; add npm test script.
Phase 1 — group ride creation (no phantom rides):
- Check authentication and Firestore availability BEFORE touching UI state; an unauthenticated
user (or unavailable Firestore) now gets an actionable error and no phantom ride, instead of a
local entry that was never saved.
- Insert the ride into UI state ONLY after batch.commit() confirms the write; on commit failure
nothing is inserted and an error is shown.
- Prevent duplicate submissions: the create modal disables its submit button while in flight, and
the hub guards with a ref.
- The join code remains the real Firestore document id (no fabricated ids).
- Extract commitGroupRide (pure, dependency-injected) so every path is unit-testable.
Phase 2 — replay playback timing:
- The elapsed clock and rider marker both derive from the same cursor, so they cannot drift.
- Manual seeks (scrub, restart, corner jump, arrow keys) go through seekTo(), which reconciles
playback timing (clears carry/last-frame) so no stale budget is applied after a seek.
- Make gap compression explicit: a caption labels the timeline as recorded elapsed time and warns
that dead time in multi-segment recordings is skipped during playback.
Tests: commitGroupRide (unauthenticated / unavailable / commit-ok / commit-failed); group-rides
hub now uses an authenticated mock with deterministic Firebase — authenticated create persists,
unauthenticated create is blocked with no phantom, a failed commit leaves no phantom, and the
create modal blocks duplicate submits.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The replay loop advanced the cursor by one sample every 50 ms / speed, assuming a uniform
20 Hz stream. Recorded telemetry is not always uniform (segment breaks and GPS-loss gaps up
to 5 s), so the rider marker drifted out of sync with the elapsed-time readout — which is
derived from real timestamps — and "1x" did not match real time across gaps.
Make playback timestamp-driven:
- Add advancePlaybackIndex(points, fromIndex, budgetMs): a pure helper that consumes a
budget of virtual recorded time, advancing by the real delta between samples so the
marker tracks the elapsed-time clock. Disconnected transitions (per
areTelemetryPointsConnected) are compressed to a single 50 ms step so playback never
stalls across multi-second dead time. No coordinates are altered.
- Drive the rAF loop with frameDelta * playbackSpeed plus carry-over, resuming from an
index ref so manual scrubs are honored.
Verified sound without changes: SHA-256-checked R2/inline telemetry load, CSV/GPX parse and
export integrity, loading/error/empty overlays, and tile-provider fallback.
Tests: advancePlaybackIndex covers uniform cadence, carry-over, gap compression, end-clamp,
and empty input. Existing replay/export/corner tests remain green.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Developer