v1.3 extends the replay with independently cached measured analysis and a separate fictional race workspace. Historical timing and approximate recorded coordinates remain distinct from hypothetical models.
These are release notes. The maintained user guide covers every current flow and its save, reset and recovery behavior.
Open Race analysis desk above the replay. Choose a focus driver, an optional comparison, and either full-race retrospective scope or completed laps up to the current replay clock.
- Lap pace plots recorded durations and trailing five-usable-lap medians within each stint. Summary cards show median pace, best usable lap and measured sector coverage. Stint slopes describe observed time changes with tyre age; they do not isolate tyre degradation from fuel, traffic, flags or weather.
- Sector comparison compares a selected lap number or each driver's best usable sectors. Missing measurements remain empty. A sum of best sectors can combine different laps and stints; it is not a recorded lap. A sector delta is not an on-track gap.
- Pit-lane timing separates lane traversal from stationary service time. Neither is net race-time loss relative to continuing on track. Older sessions may not expose stationary times.
- Per-lap data and replay jumps includes times, measured sectors, speed-trap readings, tyres and quality notes. Click a lap to return to that driver/time in the main replay.
- Export the focused driver's rows as CSV or the comparison as JSON. Exports retain missing values and source/exclusion information.
The pace summary excludes the opening lap, known pit in/out laps, unusable durations and large timing outliers. The filter cannot identify every traffic/flag effect. Full-race scope deliberately reveals future results; replay scope uses only completed laps.
Monza includes a separate bundle of 1,009 sector records. For another cached race, choose Load measured analysis. Refresh measured analysis keeps the current cache available until replacement succeeds; failures leave the old charts usable and offer retry. Missing or failed pit-lane retrieval is reported separately from usable sector data.
In Drivers & view, enable Recorded sector markers while using recorded positions. Pitwall selects the first suitable completed reference lap for the focused driver, combines its sector durations with its lap clock, and looks up approximate recorded coordinates at the first two sector ends.
The reference lap is shown beside the map. No marker appears before a suitable lap completes, when sector totals are inconsistent, when measurements are missing, or when the coordinate lookup crosses an unsupported gap. Marker preferences save with the default/named map layouts. These annotations are approximate crossing coordinates, not surveyed circuit boundaries.
Open Pace forecast backtest. Select a driver, a training branch and a later holdout of 1–15 laps. Find pit-free sample finds an eligible example; Open branch in replay returns to its recorded time.
The training set contains up to 12 recent usable laps ending at the branch, restricted to the current stint, with at least five samples required. Two baselines are evaluated:
- A constant forecast equal to the training median.
- A linear trend fitted only to those past laps, bounded to ±0.5 seconds of lap-time change per lap.
Later recorded durations enter evaluation only. Missing timing or pit in/out laps make the holdout ineligible. The graph and table report predicted/actual lap times, mean absolute error, signed mean error and cumulative lap-time error. Exported JSON includes the training laps, holdout outcomes and model limitations.
This is a pace diagnostic, not validation of a counterfactual pit strategy. Fuel, traffic, weather and flags may affect errors. A selected pit-free window is not a representative statistical benchmark, and cumulative lap-time error is not necessarily wall-clock race-time error.
In Custom circuit studio, create/load a valid circuit, then choose Circuit race sandbox. The editor and sandbox keep separate controls, and switching views preserves unsaved sandbox settings while pausing its animation.
- Configure base lap time and one to six named/coloured drivers. Each driver has a pace offset, one optional scheduled pit lap, and stationary service duration.
- Entry/exit percentages attach pit-lane endpoints to the actual rounded circuit route. Edit the orange handles by dragging or using arrow keys; click empty space to append a handle. Crossing/invalid pit routes block running and saving, and undo restores earlier configurations.
- Run sandbox, speed controls and Advance one lap exercise the deterministic simulation. Drivers enter the pit route, stop at its service midpoint, and rejoin at the configured exit. Frame-size invariance and exact service/rejoin transitions are unit tested.
- The running order uses simulated circuit progress. Grid markers start behind the line; their first crossing starts lap one.
- Save circuit and race setup persists both in this browser. Reopen the saved circuit to restore its driver/pit setup. Export sandbox session JSON exports geometry, assumptions, clock and state for inspection. Session-file import is future scope; circuit JSON import remains available in the editor.
Pit speed is fixed at 35% of each driver's synthetic track speed. This is a kinematic preview, not a physical racing model: grip, collisions, fuel, overtaking, tyre changes and car upgrades are not modeled. Historical replays never use the fictional circuit or driver settings.
The analysis worker fetches lap-sector and pit data independently of location feeds. It uses the existing bounded import queue, duplicate-job reuse, provider pacing and retry logic. Only a successful usable result updates the cache. Atomic PostgreSQL JSON updates preserve existing timing, locations and bookmarks, including concurrent analysis/location writes; snapshot mode provides equivalent merge behavior for one API process.
POST /api/sessions/{id}/analysis queues or reconnects to an import for a cached race; uncached timing returns 409. GET on that route reports job status. POST .../analysis?refresh=true queues a replacement even if analysis is cached. Active refresh status takes precedence over the old cache; failed refreshes leave that cache intact. Completed cache status survives API restarts.
The CLI command node scripts/import-race.mjs --session 9590 --analysis writes data/analysis/monza-2024.json, leaving the replay bundle intact. Startup seeds missing analysis without overwriting newer cached enrichment. Provider field meanings follow the OpenF1 reference.
node scripts/doctor.mjs checks installed runtimes, build artifacts, the database TCP port and active previews. It is read-only and never prints connection credentials.
After building the frontend/API, node scripts/preview.mjs serves both on 127.0.0.1:5080. It checks prerequisites and database reachability first, recognizes an existing Pitwall instance, accepts --port 5082, and supports temporary --snapshot storage. Ctrl+C stops its child API. Development hot reload and Docker Compose remain available.
Large tool modules load on demand. An isolated error boundary handles tool-download/runtime failures while keeping the main replay recoverable. Replay state uses immutable per-race indexes and binary searches; track-distance interpolation caches its compiled path. A manual benchmark against the v1.2.0 tag matched 1,000 replay frames and 1,000 track samples before reporting timing. See testing for results and limits.
- 61 Vitest checks, nine Node checks and 44 Chromium browser scenarios.
- Real API checks in snapshot and PostgreSQL modes, including cache/bookmark persistence across restart.
- Dedicated storage checks in both modes for independent enrichment merges, concurrent writes, downgrade protection, bundle-seed safety and bookmarks.
- Live Interlagos analysis: 1,437 sector records and 58 pit visits, with identical hashes of existing location traces, identical clock origin and a preserved verification bookmark.
- Visual review of desktop dark and mobile light layouts, plus checked-in demo images.
The release targets local master and the annotated v1.3.0 tag. Publish through the existing helper from your own signed-in PowerShell:
node scripts/push-github.mjs master feature/v1.3-labsThe push is atomic and includes reachable annotated tags. GitHub Actions must pass on the published commit before the hosted container is treated as verified. Codex's Windows session cannot persist GitHub credentials; publication uses your own terminal as with v1.2.