| Filename | Latest commit message | Latest commit date |
|---|---|---|
b0688c2a2f
Some checks failed
repository checks / lockfiles, advisories, tests in CI, workflow hygiene, spec and section references, public text (push) Has been cancelled
Revalidation (ARL 7+) now has a stated test. Each category compares the candidate's pass/fail counts with the baseline's by a one-sided Fisher exact test; Holm's step-down procedure across categories holds the family-wise error at a stated alpha, so one borderline category among many is not called a regression the way testing each at alpha would. The p-value uses only +, * and /: each hypergeometric term is reached from its neighbour by a ratio of integers, starting at the mode, so the verdict is the same bits on every platform and nothing overflows at thousands of runs. arl-core::regression (fisher_worse, holm, regression) and `arl regress <comparison.json>`, exit 0 pass / 1 regressed / 2 cannot compare. Checked against an exact u128 computation, including the N-K < n case the CLI test exposed as an underflow. Falsified: testing each category at alpha, Holm without its stop, and the upper tail each turn tests red. Co-Authored-By: Claude Opus 5.5 <[email protected]> |
||
| .. | ||
| src | ||
| Cargo.toml | ||