# Pocket-pair evidence lab: reproducibility note

Generated and checked 2026-09-03. This bundle supports the GTO Gecko article “Can Hand Histories Prove a Poker Shuffle Is Fair?”

## Files

- `poker-shuffle-pocket-pair-power.json`: null model, assumptions, exact cutoffs, power values, sources, and limitations.
- `poker-shuffle-pocket-pair-power.csv`: the same published rows in a flat format.
- `poker-shuffle-evidence-verifier.py`: independent Python standard-library checker.
- `poker-shuffle-evidence-README.md`: this note.

## Reproduce

Place all four files in one directory and run:

```text
python poker-shuffle-evidence-verifier.py
```

The verifier independently reconstructs the binomial probability mass around its mode, derives the equal-tail cutoffs, recomputes power under 4% and 8% planted rates, and checks the CSV against the JSON. It does not import or call the browser calculator.

## Declared model

For a uniformly shuffled 52-card deck, a hero pocket pair occupies `13 × C(4,2) = 78` of `C(52,2) = 1,326` unordered two-card combinations, so `p0 = 1/17`. The published rows use one outcome check fixed before collection and a 5% family error budget. Each tail receives 2.5%; because the binomial is discrete, the actual false-flag probability is generally smaller than 5%.

These calculations address only compatibility of one count with one declared model. They cannot certify randomness, identify why a count is unusual, audit source code or a deployed binary, evaluate entropy or unpredictability, or assess operator controls. A selected, incomplete, or repeatedly inspected sample does not satisfy the declared model.