Every rival is a matchup
Each recorded classification becomes a set of head-to-head results. Non-starters are excluded. Finishing ahead of a stronger rival creates a larger gain; an expected result creates less movement.
RATINGS METHODOLOGY
What the score
does—and does not—say.
A transparent guide to the evidence, assumptions and validation behind Racelytic Ratings.
MODEL VERSION
Ratings currently use Elo 1.4. Each championship has an independent rating pool and the published settings are retained unless historical validation supports a change.
THE MODEL
A form signal,
not a verdict.
Context changes the weight
Full-distance Grands Prix, E-Prix and feature races carry full weight. Sprints and reverse-grid races carry half weight. Shortened races are scaled by completed distance, while retirement matchups are softened by laps completed. Disqualifications retain full effect.
Experience controls movement
The K-factor uses three stages: 140 for the first 10 rated events, 98 through event 50 and 70 thereafter. New drivers can therefore move quickly while established careers react more steadily.
Series stay separate
Formula 1, Formula 2, Formula 3, F1 Academy and Formula E each begin from an independent 1500-point pool. The published setup keeps every event zero-sum within its pool, so ratings do not inflate through normal updates and cannot be compared directly across series.
UNCERTAINTY
Evidence, not an error bar
Everyone begins at 1500 with uncertainty of ±260. Weighted evidence narrows that range towards a ±45 floor, while inactive evidence halves every three years and displayed uncertainty is recalculated for the selected date. Evidence labels reflect each championship’s calendar: Stable begins at 50 weighted events in F1, 30 in F2 and Formula E, 24 in F3 and 16 in Academy. These labels describe how established the evidence is—not the possible error in a race prediction.
READING A PROFILE
Current form and career context
Current rating is the score after the latest rated event; peak is the highest score reached at any point. Expected converts the pre-event pairwise probabilities into an implied finishing position. Change is that event’s rating movement, while Weight reports the format and distance multiplier applied before retirement-specific matchup adjustments.
VALIDATION
Changes have to earn their place
Candidate settings are tested chronologically: the model learns from earlier races and is scored on later, unseen periods. Pairwise and equal-event accuracy, Brier score, log loss, finishing-position error and active-field rating movement are monitored. Confidence intervals resample complete race weekends so the many comparisons from one event are not treated as independent evidence. Promotion also requires consistent folds, a minimum improvement and agreement with the later-period holdout. A more complex setting is not published merely because it fits one series or era better.
LIMITS
Results include more than the driver
Car performance, qualifying pace, strategy, reliability and mechanical luck cannot be fully separated from classifications. Racelytic Ratings describe competitive outcomes against the field. They are not a pure measure of driver talent and should not be read as one.
LATEST VALIDATION
Published performance checks
Loading the latest saved validation report…
MODEL HISTORY
A versioned public record
Elo 1.4 is the current competitive model. Every ratings page displays its active model and recalculation date; saved validation reports remain tied to the model version and championship that produced them.
Current interpretation: use the rating to follow competitive form, the rank to understand position within a series, and the evidence label to judge how established that estimate is. Cross-series rating comparisons are not supported.
F1 team-adjusted beta: the optional beta separates a persistent driver component from a constructor component and ranks the combined pre-race estimate. It is validated for prediction, but the components remain model estimates rather than direct measurements of driver talent or car performance. Competitive remains the default rating.