Models / gradient-boosted-ensemble

Tyre Cliff

P(tyre performance cliff), per driver per lap.

gradient-boosted-ensembleanalytics.strategy

How it works

Cliff heads score the same feature vector as LTOE — zero extra inference cost — and are calibrated against held-out data. The output rides the LTOE payload as a sibling field, flagging when a tyre is approaching the steep part of its degradation curve rather than linear wear.

Inputs

  • LTOE feature vector
  • stint age vs compound life
  • degradation trend

Output

drivers[].pace.tyre.cliffCliff probability (p), proximity (withinLaps, lapsToCliff), and the detected flag.

Calibration

Calibrated against held-out stints. See the calibration reference for the shared pipeline.

← All models