Where the numbers come from and why they mean what they mean.
Formula 1 results tell you who finished where. DeltaBase asks a different question: given the car each driver was driving, what position should they have finished? The gap between what happened and what should have happened is the signal.
Data is sourced from FastF1's official timing data after each session. It arrives in our database within minutes of the session ending.
We start with where each driver qualified. A driver who qualified P3 in a car that belongs at P5 is already outperforming their machinery by two places. We measure that gap-to-pole — how far behind the fastest car their machinery actually is — and use it as the baseline for where a driver should finish.
We then blend in Friday's long run data. If a car that qualified P8 consistently ran race-distance laps at P5 pace, we expect it to finish around P5. The blend is 80% qualifying position signal, 20% Friday practice. The Friday data only changes the prediction when it's strong enough to trust.
If we don't have strong Friday data — sprint weekends, for instance — we fall back to the qualifying-only model.
Confidence rating: Each prediction carries a confidence level — High, Moderate, or Limited. High means we had strong practice data and clear qualifying separation. Limited means we're relying more on historical averages for that circuit.
Qualifying is measured in two layers. The first asks which session a driver reached — did they make Q2 when their car belongs in Q1? Reaching a higher qualifying zone than the car deserves is itself outperformance, before any lap time is considered. We know from the car's season-average pace roughly which zone it should reach.
The second layer measures the gap to the fastest time within the zone the driver actually reached — how competitive were they once there? These two signals are weighted equally. Where reliable, Friday's FP3 pace blends into the gap layer at 40% to sharpen the prediction.
Sprint weekends: There is no FP3 on a sprint weekend, so the Friday practice blend is suppressed. The zone and gap signals still apply — qualifying runs normally on sprint weekends — but confidence is lower without the Friday data.
The difference is simply: where they finished, minus where they were expected to finish. Positive means they beat the car. Negative means the car did more work than they did.
Albon finishing P8 when the car was expected P14 is a +6. Leclerc finishing P7 when the car was expected P4 is a −3. It is not a judgement — it is an observation.
Illustrative figures — not live data. Positive = finished above expectation. Negative = finished below.
The qualifying delta now captures two things: whether the driver reached a higher qualifying session than the car deserves, and how competitive they were within it. A driver who qualifies P4 in a car rated P7 has beaten the car in both dimensions — they made Q3 when they shouldn't have, and posted a strong time once there.
This surfaces qualifying specialists more clearly — drivers who consistently drag their car into a session it has no right being in.
DeltaBase does not claim Albon is a faster driver than Verstappen. It claims that Albon extracts more from his machinery, relative to what that machinery is capable of. These are different things.
The model does not account for safety car luck, team strategy calls, tyre allocation, or individual circuit quirks. These matter — and over a season, they partially average out. But no model is the whole story. The difference is a signal, not a verdict.
The current models are xF v3 and xQ v3. Version numbers change when the underlying calculation logic changes — not just when new data arrives. When we update the model, historical results are recalculated so all seasons are comparable.