pandas mask + sort_values is ~6.54× slower than polars filter + sort for filtering then sorting

Two operations chained, timed as one.

Result

Run	pandas mask + sort_values	polars filter + sort	Ratio
Run 1	67.34	10.71	6.29×
Run 2	72.9	11.15	6.54×
Run 3	67.95	10.3	6.59×

Spread over 3 runs: pandas mask + sort_values CV 4.4% / polars filter + sort CV 3.9% / ratio CV 2.5%

df[df['v'] > 0.5].sort_values('v')

df.filter(pl.col('v') > 0.5).sort('v')

Half the rows survive the filter, then get sorted ascending.

Sorting a column that is already sorted is much cheaper for both sides.

cd bench && python run.py --run 1 --out ../data/run1.json