Analytics

Analytics Reference

The Analytics tier injects six ML model outputs as an analytics key on timingdata, cardata, and tyrestintseries payloads. No additional API calls, no separate feed subscriptions — the fields appear automatically when your account is on the Analytics tier. Analytics enrichment is currently available for F1; models for other sports will expand as RaceHooks grows.

ℹ

Analytics enrichment requires the Analytics tier ($499/mo). Upgrade from the pricing page or the console billing page.

Enriched payload structure

Every timingdata payload gains a per-driver analytics object and a top-level overtakePairs array:

json

{
  "feed": "timingdata",
  "session": "2026 British Grand Prix — Race",
  "lap": 42,
  "data": {
    "1": {
      "position": 1,
      "lastLap": "1:29.104",
      "compound": "MEDIUM",
      "tireAge": 22,

      // ↓ injected by RaceHooks Analytics tier
      "analytics": {
        "tireHealth": 0.58,
        "tireHealthDelta": -0.04,
        "predictedLapsRemaining": 5,
        "cliffRisk": "high",
        "pitStopProbability": 0.79,
        "expectedWindowOpen": 1,
        "expectedWindowClose": 4,
        "winProbability": 0.61,
        "winProbabilityDelta": +0.08,
        "podiumProbability": 0.88,
        "safetyCarProbability": 0.14,
        "ltoe": -0.144,
        "ltoeCategory": "overperforming",
        "paceRank": 1
      }
    }
  },

  // overtake pairs are top-level, not per-driver
  "overtakePairs": [
    { "attacker": 4, "defender": 1, "p3lap": 0.11 },
    { "attacker": 14, "defender": 4, "p3lap": 0.44 }
  ]
}

Tire State Intelligence

Algorithm:Kalman Filter + Gaussian Process Regression

Per-driver tire health and degradation prediction, updated on every timing update.

Field	Type	Description
`analytics.tireHealth`	`number`	0.0–1.0. Current estimated tire health (1.0 = new). Values below 0.35 indicate the cliff zone.
`analytics.tireHealthDelta`	`number`	Rate of change per lap. Negative = degrading. Accelerating negative delta = cliff approaching.
`analytics.predictedLapsRemaining`	`integer`	Model-estimated laps before health reaches the critical threshold for this compound on this circuit.
`analytics.cliffRisk`	`string`	"low" \| "medium" \| "high" \| "critical". Derived from tireHealthDelta acceleration.

✦

The Kalman filter updates degradation rate on every timing packet, not just lap completions. Catches early degradation 3–5 laps before the team notices the delta.

Win Probability

Algorithm:Continuous Time Markov Chain (CTMC)

Win probability for every driver, updated on every timing update including safety car and incident events.

Field	Type	Description
`analytics.winProbability`	`number`	0.0–1.0. Probability of this driver winning the race from current state.
`analytics.winProbabilityDelta`	`number`	Signed change from the prior lap. Large positive spikes typically indicate safety car deployment benefitting this driver.
`analytics.podiumProbability`	`number`	P(top 3 finish). Useful for fantasy apps and broadcast narrative scoring.
`analytics.topHalfProbability`	`number`	P(top 10 finish). Useful for points threshold modelling.

✦

Computed via matrix exponentiation — analytically exact within model assumptions and fast enough to run on every timing packet, not just at lap boundaries.

Pit Stop Prediction

Algorithm:Bidirectional LSTM

Per-driver pit stop probability and expected window, updated every lap.

Field	Type	Description
`analytics.pitStopProbability`	`number`	0.0–1.0. Probability of a pit stop this lap. Track the gradient across successive laps — 0.31→0.44→0.62→0.78 is more informative than any single value.
`analytics.expectedWindowOpen`	`integer`	Laps from now until the probable pit window opens. null if probability is below threshold.
`analytics.expectedWindowClose`	`integer`	Laps from now until the probable pit window closes.
`analytics.undercut`	`object`	{ vs: driverNumber, gain: lapTimeDelta } — the driver most threatened by an undercut and the estimated position gain. null if no clear undercut candidate.

✦

Trained on 7,200+ documented pit stop events from 2020–2025. The bidirectional architecture lets the model look at both recent lap history and the expected end-of-stint pattern.

Safety Car Probability

Algorithm:LightGBM Gradient Boosted Trees

Real-time probability of safety car, virtual safety car, and red flag per lap.

Field	Type	Description
`analytics.safetyCarProbability`	`number`	0.0–1.0. Probability of a full safety car deployment within the next lap.
`analytics.vscProbability`	`number`	0.0–1.0. Virtual Safety Car probability — lower severity incidents than full SC.
`analytics.redFlagRisk`	`number`	0.0–1.0. Probability of a red flag. Low baseline, spikes materially on serious incidents.

✦

Trained on 400+ safety car events from 2018–2025. Features include field compression, recent incident rate, circuit sector characteristics, and weather. Updated every lap.

Overtake Probability

Algorithm:XGBoost with per-circuit calibration

Per-driver-pair probability of a position change within the next 3 laps. Returned as a top-level array, not per-driver.

Field	Type	Description
`overtakePairs[].attacker`	`integer`	Driver number of the car attempting to overtake.
`overtakePairs[].defender`	`integer`	Driver number of the car being challenged.
`overtakePairs[].p3lap`	`number`	0.0–1.0. Probability of attacker overtaking defender within 3 laps.

✦

Accounts for circuit overtake index — a 1-second gap on Monza carries different probability than the same gap at Monaco. Pairs within 3 seconds of each other are included, sorted descending by probability.

Pace Normalization (LTOE)

Algorithm:LightGBM Expected Lap Time Model

Normalized pace delta: actual lap time minus model-predicted lap time. Strips out tire age, fuel load, and track evolution.

Field	Type	Description
`analytics.ltoe`	`number`	Seconds (signed). Negative = faster than expected. Positive = slower. A value of -0.18 means the driver is lapping 0.18s faster than their tire state, fuel load, and circuit conditions predict.
`analytics.ltoeCategory`	`string`	"underperforming" \| "neutral" \| "overperforming". Derived from ltoe threshold by compound and circuit.
`analytics.paceRank`	`integer`	Driver's position in the field sorted by LTOE. 1 = fastest normalized pace.
`analytics.trafficAdjusted`	`boolean`	Whether the dirty-air traffic correction was applied this lap.

✦

Trained on 2.1M lap observations from 2018–2025. R² ≈ 0.94 on out-of-sample laps — when 94% of lap time variance is explained, the residual is real information. LTOE is suppressed for 2026 until 3+ races of 2026 data are available for the new era baseline.

2026 regulation adjustments

The 2026 F1 regulations introduced structural changes to car physics — turbo lag, tripled ERS power, and active aerodynamics replacing DRS. All six models required assessment:

Model	Status	Action taken
Tire Degradation	Recalibrated	Circuit parameters updated for new thermal loading from turbo lag. Conservative 50% wider confidence intervals until 4+ races of 2026 data.
Win Probability	Updated	DRS → Overtake Mode: same 1.0s gap threshold. Transition rate β₃ adjusted 0.90 → 1.10 for electric boost advantage.
Pit Stop Prediction	Adjusted	ERS clipping creates saw-tooth lap time variance that mimics cliff signatures. Decision threshold raised 0.53 → 0.60.
Safety Car	Offset applied	Base SC rate +0.10 for first 6 races of 2026 (driver adaptation period to turbo lag).
Overtake	Recalibrated	Circuit overtake indices updated for zone-independent Overtake Mode vs. DRS zone-specific activation.
LTOE	Suppressed	Returns null until 3+ races of 2026 data available. 2026 is a structural break — car baseline resets completely.

← Feed catalog API Reference →