We show
our work.
Most predictions tell you what. We show you why — the data behind every pick, the confidence level, and whether it held up. Accuracy is public, unedited, and scored against official results.
Ten forecast products, every race weekend.
Predicted winner, full podium, top-10 order, fastest lap candidate, and key swing factors. First published two weeks out, then refreshed after every session — the final version is generated after qualifying sets the grid.
Pole prediction, front-row probability, and drivers likely to over- or underperform their car's pace. Updated after each practice session, locked before Q1 begins.
Separate FP1, FP2, and FP3 forecasts — who tops the timesheets, who runs race sims, and whose headline lap is fuel-corrected fiction. On sprint weekends only FP1 is predicted, because that is all the practice anyone gets.
Teammate battles predicted separately for every session — practice, qualifying, sprint, and race. Who has the edge in qualifying trim, who holds the race pace advantage, and where the gap is closest.
Separate top-8 forecast for sprint weekends. Shorter distance means less strategy variance, so the model weights raw pace and grid position more heavily.
Predicted sprint qualifying order. One-lap pressure with minimal practice — the model factors in circuit character and each driver's single-lap form.
8–10 yes/no prop questions per race — safety car count, first-lap incidents, midfield podiums, driver-specific calls. Each carries a predicted answer and a probability, and every one is scored against the official result.
Championship-scale calls: WDC and WCC outcomes, season totals like race win counts, and milestone questions. Probabilities are updated as the season develops and resolved race by race.
Two or three tongue-in-cheek, circuit-specific props per race — local wildlife on track, broadcast traditions, podium theatrics. Each comes with a confidence score, and yes, they get resolved after the race like everything else.
After qualifying, sprint sessions, and the race, the AI writes a grounded debrief of what actually happened — where the forecast held, where it broke, and why.
From chequered flag to lights out.
A prediction is not one pre-race snapshot. The forecast cycle starts as soon as the previous race is scored and re-runs after every session — and every version is kept, so you can watch the picture sharpen as real data replaces speculation.
The first look at the next Grand Prix is generated as soon as the previous weekend is fully processed — often two weeks out. Pure form and circuit history, before a single wheel turns.
A fresh forecast built the day before the weekend starts. This pass searches for inter-race news — upgrades, penalties, silly-season noise — and folds it into the picture.
Every completed practice session triggers a re-run. Long-run pace, fuel-corrected deltas, and weather trends replace pre-weekend guesswork, session by session.
Once qualifying (and the sprint, on sprint weekends) is classified, the final race prediction is generated with the actual starting grid — the last update before lights out.
Every version is immutable the moment it is published — enforced at the database level. Earlier snapshots are never overwritten, and the final one is what gets scored.
Inside F1 Predictions AI.
F1 Predictions AI is not a lookup table or a points projection. It reads session data, circuit history, and current form, then reasons through the evidence to produce a structured forecast with a confidence score you can challenge.
After each session, finishing positions, lap times, grid slots, fastest laps, and weather conditions are pulled from live timing feeds. The model works from the same numbers a pitwall strategist sees.
Circuit history, championship standings, recent form, and weather data are layered in. A street circuit weekend produces a fundamentally different prediction to a power track.
F1 Predictions AI identifies the likely outcome, flags upset candidates, and writes the reasoning behind every pick. You can read the full logic chain — not just the result.
Once published, predictions cannot be edited — enforced at the database level. When the session starts, what was written is what gets scored.
The scoreboard is public. Always.
Every prediction is scored against official FIA results, race by race. Accuracy numbers are public and unedited — same scoreboard for everyone, no cherry-picking.
The predicted race winner must finish P1. No partial credit, no “close enough”. Either the winner was called correctly or it wasn't.
An overall accuracy score of 0.6 or higher across the top-3 positions. The scoring weights position accuracy, not just presence — P1 correct counts for more than P3.
Accuracy is calculated after the official race classification is confirmed. Retirements and post-race penalties are reflected in the final score.
We do not retroactively edit predictions after results are known. We do not skip scoring a race because the model performed poorly. Every round is on the board.
What F1 Predictions AI cannot do.
F1 has too many variables for any model to get right every time. F1 Predictions AI works from available evidence — here is what falls outside that scope.
Mechanical failures — a car leading by 20 seconds is still a DNF risk. Reliability data is internal to teams and invisible to any external model.
First-lap incidents — grid position matters, but contact at Turn 1 is not foreseeable from timing data.
Safety car timing — the model estimates SC probability from circuit history, but not which lap it falls on. Lap timing often decides the race.
Mid-race weather shifts — forecast data has uncertainty. Predictions made under a changing sky are the model's weakest area.
Team orders — a driver told to hold position will look slower than their actual pace. The model cannot see team radio.
Winner picks, podium forecasts, and full reasoning for every session.