Formula 1 · Telemetry intelligence
Read the telemetry.
Model the tyre.
Call the race.
LapBox loads real session data, predicts lap time and tyre degradation, and solves strategy as a duel between two cars that both get to react — then shows its working on every call.
- 01For the grandstand
One number that matters and a sentence explaining it. No stint tables, no jargon.
- 02For the pit wall
Every lap, stint and delta, fuel-corrected, with the undercut window drawn live.
- 03For the model
Training runs, loss curves and SHAP attributions behind every call the system makes.
- VER1:31.204−0.182
- NOR1:31.386+0.182
- LEC1:31.559+0.355
- Track temp41.2 °C
- PIA1:31.702+0.498
- UndercutL33–L36open
- HAM1:31.884+0.680
- Air27.8 °C
- RUS1:32.011+0.807
- SC risk18%by L38
Built on
FastF1
Session & telemetry source
XGBoost
Lap time · degradation
SHAP
Attribution
FastAPI
Typed REST surface
PostgreSQL
Processed laps
Next.js
This interface
One session, three readings
Same data. Different lens.
02 · Engineer
Everything, at pit-wall density.
Lap-by-lap pace with fuel burn corrected out, live stint state, the undercut window as it opens and closes, and the strategy call re-scored against the car you are actually racing.
- Fuel-corrected pace, dirty air priced per driver
- Undercut and overcut windows drawn as they move
- Every call audited against the rival re-planning too
Fuel-corrected pace · last 12 laps
- Gap ahead
- +2.41s
- Gap behind
- −1.08s
- Tyre age
- 14laps
- Deg rate
- 0.07s/lap
Undercut window open — laps 33 to 36
The pipeline
Four steps between a raw session and a decision you can defend.
- 01
Ingest
A whole session at once — laps, telemetry channels, weather, tyre stints — cleaned of in-laps, out-laps and anything the timing loop got wrong.
- 02
Model
Lap time and per-compound degradation, with fuel burn corrected out and dirty air priced per driver, so pace means pace.
- 03
Decide
Strategy solved as a duel: two coupled cars, every pit-wall reflex answered at its best, weighed across every lap a caution could land on.
- 04
Render
The call, the margin it wins by, and the SHAP attribution behind it — in whichever of the three lenses you are reading in.
- Modules
- 17
- Collection through documentation
- Tests green
- 934
- 719 backend · 215 frontend
- API endpoints
- 51
- Typed, across 17 routers
- Dashboard views
- 15
- Telemetry to explainability
Access
Bring the pit wall’s view to your screen.
Real sessions, real telemetry, and a model that shows its working. Request access and pick the lens that fits how you watch.