The gate's stratified interval holds nominal at both Holm levels, in H2's regime¶
scripts/crossed_coverage.py --strata 2 simulates the interval the decomposition gate reads: two
equally weighted strata, changes resampled within each, five seeds, 4,000 trials a cell (Monte
Carlo error about 0.002 at the 0.0125 level). One-sided false-positive rate, the lower bound above
zero under a true null, beside the unstratified crossed interval on the same runs. Generated from
stratified-coverage-0.975.json and stratified-coverage-0.95.json.
| level | nominal | sigma_b | stratified | crossed |
|---|---|---|---|---|
| 0.975 | 0.0125 | 0.0 | 0.0105 | 0.0100 |
| 0.975 | 0.0125 | 0.005 | 0.0112 | 0.0110 |
| 0.975 | 0.0125 | 0.01 | 0.0145 | 0.0150 |
| 0.975 | 0.0125 | 0.02 | 0.0132 | 0.0130 |
| 0.95 | 0.025 | 0.0 | 0.0222 | 0.0208 |
| 0.95 | 0.025 | 0.005 | 0.0230 | 0.0222 |
| 0.95 | 0.025 | 0.01 | 0.0238 | 0.0230 |
| 0.95 | 0.025 | 0.02 | 0.0260 | 0.0257 |
Within Monte Carlo error of nominal at both levels up to a seed effect of 0.02. These runs are
H2's regime (equal halves, the same treatment arm in both), as the re-review pointed out. H1's
second half swaps treatment and control, so an adapter-level seed shift enters the two halves
with opposite signs; that regime (--flip-second) is measured next, before the report states a
coverage for H1.