We report Wilson intervals and a McNemar p-value, but not a standardized effect size. Cohen's h is a small,
self-contained addition that helps readers judge how large a pass-rate difference is, not just whether it's
significant.
Add cohens_h(p1, p2) to tjbench/stats.py beside the existing functions (pure Python, no scipy), and
surface it in the stats block. Add unit tests for known values in tests/test_stats.py.
Acceptance criteria: cohens_h implemented + unit-tested; matches reference values for a few known
proportions.
We report Wilson intervals and a McNemar p-value, but not a standardized effect size. Cohen's h is a small,
self-contained addition that helps readers judge how large a pass-rate difference is, not just whether it's
significant.
Add
cohens_h(p1, p2)totjbench/stats.pybeside the existing functions (pure Python, no scipy), andsurface it in the stats block. Add unit tests for known values in
tests/test_stats.py.Acceptance criteria:
cohens_himplemented + unit-tested; matches reference values for a few knownproportions.