Fit a correlated race

Six runners in two camps; each camp shares a luck factor. Drag the observed win shares and the correlation, and the engine inverts the shares to standalone abilities live in your browser — the same calibration the Python package ships, at machine-precision parity. The curves are the fitted performance densities (switch the runners to Gumbel and they skew). Below them, hollow dots pretend the runners are independent and filled dots admit the correlation: a runner whose camp is strong is better than its share suggests.