Kalman Skill

A state-space model of each competitor's latent skill, per event (tabs above the table), in log-time, tracked weekly with a Kalman filter and RTS smoother (the exact linear-Gaussian equivalent of TrueSkill-Through-Time). Skill is a damped local linear trend — a level plus an improvement rate that bends toward a plateau — with robust (Student-t) handling of outlier weeks and a coupled filter tracking each person's solve-to-solve spread (CV). Unlike the EWMA Skill Estimator, this carries uncertainty, so a Monte-Carlo ao5 simulator gives expected ao5, P(sub-X), and record probabilities. Skill (single) is the smoothed mean solve time projected to the current week; Trend is the weekly improvement rate (green = improving).