We use the Bradley-Terry model, for all our true score and infit calculations and aim to be consistent with the btm function in the R package sirt. We have open-sourced the code, and the algorithms are described in more detail, on npm. We welcome contributors!