pub fn apply_percentile(stats: &mut [Stats])vcs-git only.Expand description
Recompute every file’s risk_score as the mean percentile rank of
its signals within the analyzed set (the --risk-formula percentile
mode).
Each signal is ranked independently using the mid-rank of ties
(so identical values share a percentile), scaled to [0, 100], and
the per-file mean across signals becomes the score. With fewer than
two files percentiles are undefined, so the weighted score is left
in place.
The u64 → f64 signal casts are exact for every realistic churn
count (well under 2^53) and the result is ordinal, so the precision
lint is allowed for the whole pass.
The resulting risk_score is stamped with and cache-keyed on
RISK_SCORE_VERSION (the same constant as the weighted formula),
so any change to this blend — the extractor set, the mid-rank
scaling, or the / signal_count * 100 normalization — must
increment that constant.