# Poker study confidence exercise

Original teaching material for [Poker Study Confidence: Check Your Sure Answers](https://gtogecko.com/blog/poker-study-confidence). Sources checked September 8, 2026.

Open `exercise.html` in a browser. It is self-contained: no installation, account, network request or external script is required for the exercise itself. Source links open external pages. Answers and confidence ratings are held only in the open page; reloading clears them. Copy your notes before leaving. JavaScript records all eight answers before revealing feedback; without it, use paper and reveal the manual answer key afterward. Printing provides a question sheet and key. Answer keys exist in the downloadable source: this is a study aid, not a secure examination.

Confidence is the judged probability that the selected True/False answer matches the stated key. Choices from 50% to 100% in ten-point steps are editorial. A missing choice is not zero, a guess or a wrong answer. The 80% split groups records for review; it is not a validated cutoff. All four correctness/confidence groups remain visible.

## What the example establishes

The sixteen comparison rows are invented, not participant data. A and B each have six correct answers out of eight, four 90% judgments and four 60% judgments. Each therefore has 75% accuracy and 75% average confidence. A has two incorrect 90% answers; B has none. Equal aggregate accuracy and confidence can conceal different item-level pairings. Neither record estimates calibration or ranks learner ability.

No simulation, randomness or seed is involved. No strategic EV model or real learner experiment is supplied. The eight selected questions concern objective rules and one closing-call calculation; they are not a validated assessment. Retakes must not be counted as fresh independent observations. No gambling or educational outcome is predicted.

## Reproduce and check

Download all files in this directory together: `inputs.json`, `engine.js`, `exercise-ui.js`, `generate.py`, `verify.py`, `comparison.csv`, `comparison.json`, `exercise.html`, this README and `SHA256SUMS`.

With Python 3.11 or later (standard library only):

```text
python generate.py --output regenerated
python verify.py
python -O verify.py
```

The generator reads `inputs.json`, embeds both local JavaScript files, and writes the comparison CSV/JSON and standalone exercise. The verifier independently checks the source-audited answer-key sequence, exact rational call arithmetic (including a tie witness), starting effective-stack identity, grouped counts and byte-identical regeneration in a temporary directory. Browser interaction and keyboard/render checks are documented in the publication report; the Python check does not emulate a browser.

CSV fields: `record` is A or B; `item` runs 1–8 within that record; `correct` is 1 for a correct answer and 0 for an incorrect answer; `confidence_percent` is confidence in that selected answer. These abstract synthetic grades are not actual attempts at the downloadable quiz. The JSON supplies the equivalent group summaries.

The question sources and explanations are in `inputs.json` and the exercise key. Hand rules use standard 52-card Hold'em except the explicitly named four-card Omaha item. The river calculation has no rake, side pot or later wager; expected pot share includes ties. Effective stacks are measured before compulsory posts and without a separate betting cap.

The reference papers concern general-knowledge learning, not this poker exercise: [Butler, Karpicke and Roediger (2008)](https://learninglab.psych.purdue.edu/downloads/2008/2008_Butler_Karpicke_Roediger_JEPLMC.pdf), and [Butler, Fazio and Marsh (2011), abstract](https://link.springer.com/article/10.3758/s13423-011-0173-y). No papers, private app code, user data or campaign data are redistributed here.