Models / gradient-boosted-ensemble

Tyre Cliff

P(tyre performance cliff within the next few laps), per driver per lap, plus live tyre health.

How it works

Two layers work together. A tyre-health tracker follows each set lap by lap and estimates its health, degradation rate in seconds per lap, predicted cliff lap and remaining useful life. Alongside it, a learned cliff head scores the same per-lap inputs as the LTOE pace model and returns a calibrated probability that the tyre falls off its cliff — the steep part of the degradation curve rather than linear wear — within a stated number of laps. Intermediate and wet tyres use a fixed rule instead of the learned model.

When to use it

• Show viewers how many laps are realistically left on a set of tyres.
• Anticipate a pit stop or a defensive problem before lap times show it.
• Model tyre strategy in a game or simulation with a live degradation and cliff signal.

Inputs

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

Output

Example payload

An excerpt of one analytics.strategy delivery, trimmed to the fields this model produces. Values are illustrative. The full message shape is on the feed reference.

{
  "feed": "analytics.strategy",
  "lap": 35,
  "drivers": [
    {
      "number": "4",
      "tla": "NOR",
      "pace": {
        "tyre": {
          "health": 0.58,
          "healthPct": 58,
          "degRateSecPerLap": 0.07,
          "stintLaps": 19,
          "cliffLapPredicted": 41,
          "remainingUsefulLife": 6,
          "cliffRisk": "NONE",
          "cliff": {
            "p": 0.22,
            "withinLaps": 5,
            "detected": false,
            "lapsToCliff": null
          }
        }
      }
    }
  ]
}

Use it

Subscribe to analytics.strategy with a webhook and read this model's fields from each delivery:

curl -X POST https://api.racehooks.io/v1/webhooks \
  -H "Authorization: Bearer $RACEHOOKS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"webhookUrl":"https://your-app.com/f1","feedId":"analytics.strategy"}'

To try it without waiting for a race weekend, replay the recorded Canadian Grand Prix (2026-canada_r) from the quickstart.

How to read it

cliff.p is the forward-looking signal: the chance of a cliff within cliff.withinLaps laps. Read it with remainingUsefulLife and degRateSecPerLap — a rising degradation rate with a growing cliff probability is the pattern before a stop. health is a gradual wear reading and can still look mid-range at a real cliff, so do not wait for it to reach zero. cliffRisk describes the kind of wear currently visible.

Limitations

• The blistering readouts in cliffRisk describe what is happening now; they are not an advance warning.
• health can read mid-range when a cliff actually arrives — use the cliff fields for timing.
• Intermediate and wet tyres use a fixed rule rather than the learned model.
• Published in Race and Sprint sessions only.

Calibration

The cliff probability is calibrated against held-out stints. See the calibration reference for the shared pipeline.

Related models

• Pace Normalization (LTOE) — Lap Time Over Expected — the normalized pace delta, per lap.
• Pace Mode — Which pace mode a driver is running — push, hold, manage, warm-up or delta — per lap.
• Pit Stop Prediction — Probability of a green-flag pit stop on the next lap; stops under safety car, VSC or red flag are not counted.
• Tire Compound — The next compound a pitting driver takes, as probabilities.

Get the data

This model is delivered on analytics.strategy. Subscribe by webhook or stream it over SSE. The quickstart walks through registering an endpoint and replaying a past session against it.