Circuit Paul Ricard

Average absolute error on 624 classified finishes in 36 earlier races within the last ten years. Each race was held out and predicted using only older races; all three models use the same samples.

Elo + qualifying

2.14places

Qualifying only

2.38places

Elo only

2.62places

The range scales the 80th percentile of earlier held-out absolute errors to this field size, rounding outward. It is an empirical error guide, not a guaranteed 80% chance for a particular driver. We require at least five checked races and 100 checked finishes before showing it.

Inputs are field-normalised Elo rank and qualifying position. The target is position among all race entries, trained only on classified finishers. Estimates are averages, not a unique predicted finishing order. Penalties, the circuit, weather, incidents and strategy aren't model inputs. Historical qualifying coverage is patchy; this is an exploratory backtest on corrected records, not proof of live forecasting accuracy.