ETH 2023-01-15..2026-09-02 (1325 days). Null = skip the same number of gate60-long days at random. N perms = 4000. Rule: REJECT if p>=0.05.
| Variant | n_long | n_skip | base SR | gated SR | obs lift | null mean lift | null p95 | p | verdict |
|---|---|---|---|---|---|---|---|---|---|
| skip if z24>0.5 | 366 | 108 | 1.05 | 2.48 | +1.43 | -0.06 | +0.09 | 0.000 | SIGNAL |
| skip if z24>0.0 | 366 | 178 | 1.05 | 2.88 | +1.83 | -0.11 | +0.07 | 0.000 | SIGNAL |
Verdict (PRIMARY skip z24>+0.5): SIGNAL
- REJECT -> the depth z-skip adds nothing over skipping the same number of long days at random; the overlay's lift is reduced exposure, not depth information.
- SIGNAL -> depth timing beats random skipping.