# Reproducibility bundle

Working paper v0.1, "Decision Receipts: A Verifiable Primitive for AI Evidence
That Survives a Challenge".

These are the scripts behind the numbers in the paper. They take no
credentials and reach no network, so they run as-is.

| Script | Produces |
| --- | --- |
| `bootstrap_w19.py` | The W19 sample-bias bootstrap. Writes `bootstrap_w19_results.json` and `../figures/w19_bootstrap_distribution.png`. |
| `benchmark_signing.py` | Signing overhead measurements. Writes `benchmark_signing_results.json`. |
| `figure_w19_timeline.py` | `../figures/w19_timeline.png` |
| `figure_table1_heatmap.py` | `../figures/table1_heatmap.png` |

## The W19 bootstrap

Section 5 reports p = 0.222 for the W19 result. That figure is the probability
of placing the WHO-confirmed tile at rank 4 or better under a uniform-random
null over the 18 active tiles, not a claim of predictive capability. The script
re-runs the simulation at three base rates (0.1%, 0.4%, 1%) so the result is
reported as a range rather than a single number resting on one unverified
assumption.

```
python3 bootstrap_w19.py
```

Requires Python 3.9 or later, plus `numpy` and `matplotlib` for the figure.