Foundation
01What Apex Pulse predicts
The core output is each driver’s predicted gap to pole and the resulting qualifying ranking. The public artifacts do not currently claim pole or Q3 probabilities.
Methodology and trust
A pre-qualifying prediction with a clear evidence trail.
Apex Pulse turns public practice-session evidence into a qualifying forecast, freezes it before qualifying, and evaluates it only after official results are available.
Prediction foundation
A focused output, built from public evidence.
Foundation
02Data sources
New sessions use OpenF1’s free historical laps, sectors, tyres and race-control messages after its 30-minute access delay. Earlier forecasts and training data retain their original FastF1 source. FastF1 still supplies the calendar. OpenF1 lap-quality checks use a documented consistency proxy and can differ from FastF1’s validation.
How Apex Pulse works
One controlled lifecycle from practice to evaluation.
- 03
Practice checkpoints
The dataset supports predictions after FP1, FP2 and FP3. Each checkpoint can use only the sessions that have already happened.
- 04
Feature engineering
Interpretable practice pace, sector performance, tyre context, teammate-relative pace, session conditions and strictly time-aware historical form are the foundation.
- 05
Model policy
Candidate tabular models are compared with strong practice and historical baselines using season-aware, walk-forward evaluation. The selected policy is exported by the backend.
- 06
Forecast and settlement
The forecast is written before qualifying and preserved immutably. Settlement is a separate post-qualifying comparison; it never rewrites or fills missing predictions.
Evaluation and transparency
Performance claims stay attached to their evidence.
07
Evaluation metrics
Gap MAE and RMSE measure pace error. Mean absolute position error and top-k agreement show how well the ranking matched the official order.
08
Public-data limitations
The project has no private team data, paid live feed, fuel state, brake pressure, battery state or real-time telemetry. Public signals are partial and may be stale or incomplete.
09
Prospective evidence and backtests
Predictions made before an observed qualifying session are reported separately from historical model-development backtests and legacy noncanonical records.
Technical lifecycle details
- Contract version
- Dashboard schema 1.0
- Access mode
- Read-only exported JSON
- Forecast preservation
- Immutable snapshots
- Schedule source
- Existing local FastF1 cache
- Operator actions
- Not exposed by the website
- Live telemetry
- Not provided