About & Methodology

A self-hosted F1 telemetry and ML prediction platform covering the full 2026 season. No third-party ML services, no LLM commentary — every number is derived from FastF1 and OpenF1 raw data by models running on the same server as the API.

The platform currently surfaces 14 distinct analytical views. All of the following are public — no sign-in required to read them:

Full session drill-down (lap telemetry, FP intelligence panels, race intelligence panels, and session DNA) requires signing in. The dashboard is the main entry point for per-event and per-session analysis.

Most F1 analysis tools show you what happened. This one commits to a prediction before the session starts, then shows you how right or wrong it was.

Pre-session predictions are locked with a write-once sentinel the moment FP3 completes. No post-session data can alter a locked prediction — the locked timestamp is recorded alongside each artifact so the claim is permanently verifiable. The running Spearman ρ on the homepage is the cumulative season score across every published race prediction. No competitor exposes calibration this publicly.

The platform also produces a driver DNA profile across ten performance dimensions (tyre management, qualifying ceiling, race start, degradation resistance, teammate edge, overtake rate, consistency, ERS efficiency, pit execution, and racecraft) that accumulates across the full season — not just a single weekend snapshot.

Telemetry and timing data come from FastF1 and OpenF1. All sessions are normalised so laps, drivers, and years are directly comparable across the entire dataset.

Each session also ingests the full weather record (air temp, track temp, humidity, pressure, rainfall, wind speed and direction), sector split times, and the race control message log. These feed the live intelligence layer and the post-session narrative alongside the lap data.

Sessions are ingested after they finish, not during.

Practice data feeds a multi-stage model. FP1 establishes pace baselines. FP2 calibrates tyre behaviour. FP3 finalises the prediction and locks the artifact.

Once FP3 completes, the race and qualifying predictions are locked with a write-once sentinel. Back-filling after the result is known would defeat the entire credibility purpose of the platform.

Sprint weekends have only one practice session. FP1 carries the full analytical load: all pipeline stages run from that single session rather than across three. Sprint Qualifying predictions are locked from FP1 alone.

After the session, the actual result is loaded and the prediction is scored automatically — no manual intervention.

Practice sessions produce several analytical outputs beyond the base pace model, all surfaced via the FP Intel tab on each session page.

Long-run pace — Representative pace extracted per compound per driver from sustained long-run sequences, shown as a sortable table on the FP1/FP2 session page.

Track evolution — Lap-over-lap grip build-up is modelled across the session. FP1/FP2 show a single-session view; FP3 shows a multi-session comparison so you can see how rubber accumulated across the full practice programme.

Weekend progression — Each driver's best lap per session (FP1, FP2, FP3) is compared, with the delta highlighted on the event page. Improvement from FP1 to FP3 can reflect setup gains, fuel load differences, or deliberate sandbagging.

Tyre selection asymmetry — The fleet compound distribution (how many laps each compound was run across all cars) is shown alongside a per-driver table flagging outliers who deviated from the field median. A driver on a different compound than the rest of the field is a meaningful strategic signal heading into qualifying.

Teammate delta (each pairing on their best clean lap of the session) and the compound testing matrix (which compounds each driver covered and how many laps) are also derived from practice data and feed the race prediction pipeline.

Race and Sprint sessions surface four dedicated panels on the Race Intel tab of the session page. All four are public and absent-tolerant — they render nothing when the artifact has not yet been computed.

Overtake analysis — Detects on-track position changes lap-by-lap and classifies each pass by likely cause. Pit-lane passes are excluded automatically.

Driver of the Day — A composite five-component score per driver: pace performance (consistency vs. teammate and field average), stint management, overtake contribution, energy efficiency, and pit execution. Scores are normalised 0–100 within each race.

Pit window compliance — Compares the pre-race predicted pit window for each driver against the actual stop. Each driver gets a verdict badge: On-window, Early, Late, or Undercut.

Stint reconstruction — A full per-driver compound stint sequence: compound, start lap, and lap count. Colour-coded by compound (SOFT red, MEDIUM yellow, HARD white).

Driver DNA — Each driver has a season-aggregate radar profile computed across ten performance dimensions from FP and race data. Profiles accumulate across the full season, not just one event. The Driver DNA page shows all 2026 drivers ranked by their top dimension.

Head-to-Head Rivalry — Any two drivers can be compared season-long at /rivalry/2026 using their three-letter abbreviation. The page shows a DNA radar overlay and a per-event head-to-head finishing position and sector gap table.

Championship Projection — Updated after each race, the championship page shows ML-driven title probability and projected final points (p10/p50/p90 range) for every driver and constructor, updated after each race.

Season Map — The season map shows predicted vs actual finishing positions in a heat-map grid across all 2026 events. Green cells = model called it correctly; red = the model was wrong. Useful for spotting which circuits the model consistently struggles with.

During an active session, the live timing panel updates every 15 seconds with per-driver intelligence derived from the laps seen so far. It is not a lap-by-lap ticker — it is a strategy layer built on top of real data.

Each driver row shows the current compound and tyre age, the estimated degradation rate in seconds per lap, and a recommended pit window calculated from the fitted degradation model. An undercut flag marks drivers where pitting one lap early would cost the car behind the position on raw pace.

A track status banner appears above the table when a Safety Car, Virtual Safety Car, or SC-Ending phase is declared via the race control feed. During a Safety Car or VSC, the undercut and pit-window recommendations are suppressed — they are meaningless under neutralised conditions and showing them would be misleading.

The weather strip shows live conditions: air and track temperature, humidity, pressure, rainfall, and wind. Track temperature drives tyre warm-up time and is the most tactically relevant figure during tyre changes.

For races, ranking accuracy is measured with Spearman rank correlation (ρ) — a number from −1 to +1 that describes how closely the predicted finishing order matched the real one. A ρ of 1.0 means the order was perfect; 0 means no better than a random guess. The running season ρ on the homepage is the average across all scored races in 2026.

Two baselines are published alongside the model score: the grid-order baseline (predicting drivers finish where they start) and the championship-standing baseline (predicting drivers finish in their current points order). Beating both baselines consistently is the minimum bar for the model to be considered meaningful.

For qualifying, error is measured with MAE (mean absolute error) — the average gap in positions or lap-time delta between predicted and actual. Pit stop timing is evaluated separately: what fraction of predicted pit windows were within two laps of the actual stop.

The credibility page shows the full history: per-round ρ, drift index, and calibration charts across the entire season. The accuracy scoreboard shows the same data in a compact tabular form with both baselines.

Raw lap times include fuel load and tyre degradation. Direct comparison conflates strategy with speed: a driver on fresh tyres with light fuel looks faster than a driver managing degradation on a heavy load, even if their underlying pace is identical.

Raw laps are adjusted for fuel weight and tyre age to allow direct comparison across stints and sessions.

After both corrections, the True Pace table ranks drivers by corrected median pace. Corrections are only as good as the degradation model — drivers who did not complete representative long runs carry higher uncertainty. SC and VSC laps are excluded from the correction input entirely.

Early in the season — or for a new circuit — confidence is intentionally capped at 45%. There simply is not enough prior data yet to justify a stronger claim, so the system refuses to pretend otherwise. Confidence rises naturally as more events accumulate.

Overtake counts rely on position data aggregated to lap-boundary precision. Coverage varies across sessions — some have sparse OpenF1 position updates, leading to lower detected counts than the actual on-track figure.

Live intelligence is session-incremental, not real-time. Pit window recommendations and degradation rates update as each completed lap is ingested — typically within a few seconds of lap completion — but are not continuous streams. There is no real-time ingestion capability.

This platform does not use external services for ML inference, LLM-generated commentary, or third-party enrichment. Every number shown is derived from the raw FastF1 and OpenF1 data by models running on the same server as the API.

The FastF1 HTTP cache was created 1 May 2026. Pre-May sessions cannot be backfilled — the livetiming mirror does not serve old practice session data.