Trust / conversion

How We Decide Something Is Actually a Poker Leak

A leak is declared only after evidence, opportunities, and confidence are on the table. Insufficient evidence is a valid result.

  1. 1. Parse the hand correctly

    This step stays in the domain contract so a future analyzer can implement it without changing the site.

  2. 2. Normalize the decision context

    This step stays in the domain contract so a future analyzer can implement it without changing the site.

  3. 3. Detect candidate patterns

    This step stays in the domain contract so a future analyzer can implement it without changing the site.

  4. 4. Count relevant opportunities

    This step stays in the domain contract so a future analyzer can implement it without changing the site.

  5. 5. Check sample reliability

    This step stays in the domain contract so a future analyzer can implement it without changing the site.

  6. 6. Inspect decision composition

    This step stays in the domain contract so a future analyzer can implement it without changing the site.

  7. 7. Compare with strategy evidence

    This step stays in the domain contract so a future analyzer can implement it without changing the site.

  8. 8. Estimate potential impact

    This step stays in the domain contract so a future analyzer can implement it without changing the site.

  9. 9. Assign confidence

    This step stays in the domain contract so a future analyzer can implement it without changing the site.

  10. 10. Prioritize by value of fixing

    This step stays in the domain contract so a future analyzer can implement it without changing the site.

  11. 11. Re-test on future hands

    This step stays in the domain contract so a future analyzer can implement it without changing the site.

  12. What AI does — and what it doesn't

    AI explains evidence. It does not invent hand actions, stacks, samples, benchmarks, or solver outputs.

Questions

Does the AI invent solver outputs?
No. AI may explain evidence that already exists. It must not invent actions, stacks, samples, benchmarks, or solver numbers.
What if the sample is too small?
The review should say insufficient evidence and refuse a fake conclusion.

See the methodology applied to a sample.

Use the database you already built. Post-session analysis only.

See sample analysis