Rider rating · All races
μ · normal
Top 10 · at snapshot
Partial-pairing Bradley–Terry model with calendar-time σ inflation and race-importance τ scaling. Each rider has a (μ, σ) at every point in time across 8 discipline tracks, plus a cross-discipline cumulative score.
P(A beats B) = Φ( (μA − μB) / √(2β² + σA² + σB²) ).
σ already enters the denominator, so the probability is
independent of the safe/normal toggle — it reflects the model's
genuine uncertainty. β = —.
Evaluated live at the selected snapshot for any tracked rider.