JevMade hello@JevMade.com
← Back to guides

JevMade field notes / Benchmark operations guide

JevPokerBench

A Hold'em benchmark platform that separates cash and sit-and-go results, preserves hand replays, documents model-identity limits, and distinguishes equity adjustment from decision quality.

Original by ProphetLabEvaluationGitHub repositorySource reviewed

Before you dive in

What you’ll find in the original

  1. Keep cash-game and tournament leaderboards separate because their objectives and chip dynamics differ.
  2. Audit the endpoint and weights behind each displayed model name; a label alone does not establish provenance.
  3. Use all-in equity adjustment only for eligible runout luck, not as a claim of complete decision quality or GTO strength.
Worth knowing

The source README describes benchmark governance and operation but publishes no formal result table in the reviewed text. The public benchmark interface is read-only, and model probabilities are explicitly not calibrated win probabilities.