# Post-hoc: does the verdict survive reed-agent's bootstrap?

NOT part of the seal. Run on the same data as `analyse.py`. `python research/rain-before-seven/reed-bootstrap.py`.

The point estimates are unchanged by any of this — only the interval is at issue. Pooled **-33.39** points (the sealed statistic, which its own null control voided at -3.87 on information-free labels); stratified **-31.85** points, which is the one that carries the published verdict.

## The stratified statistic, which carries the verdict

| resampling unit | n units | 95 % CI | SD | verdict at the sealed bar (CI upper < -5) |
|---|---|---|---|---|
| (site, calendar month) — **the sealed one** | 1872 | -32.74 to -31.01 | 0.4398 (1.00x) | **KILLED** |
| calendar month, all six sites together | 312 | -32.99 to -30.74 | 0.5785 (1.32x) | **KILLED** |
| calendar year, all six sites together | 26 | -33.31 to -30.46 | 0.7308 (1.66x) | **KILLED** |

## The pooled statistic, void, shown for completeness

| resampling unit | n units | 95 % CI | SD | verdict at the sealed bar (CI upper < -5) |
|---|---|---|---|---|
| (site, calendar month) — **the sealed one** | 1872 | -34.24 to -32.59 | 0.4313 (1.00x) | **KILLED** |
| calendar month, all six sites together | 312 | -34.50 to -32.32 | 0.5554 (1.29x) | **KILLED** |
| calendar year, all six sites together | 26 | -34.80 to -31.97 | 0.7257 (1.68x) | **KILLED** |