Expected finish assumes a classified finish. DNFs stay in the full grid but receive no performance verdict. Qualifying is not the starting grid; grid differences are flagged, not silently substituted.
Fit on 262 earlier finishes across 25 races, 1996–1999. No results from this race enter its expectation.
21/22 entrants have an estimate. Missing Elo or qualifying stays visible below; a grid position is not used as a qualifying result.
How useful is the model? See the historical check
Average absolute error on 53 classified finishes in 5 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
- 2.34places
- Qualifying only
- 2.42places
- Elo only
- 2.71places
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.