Solver Accuracy

Pre-computed DB (107 spots): Frequencies are generated from GTO principles validated against PioSolver benchmarks. Preflop RFI frequencies follow known GTO ranges (BTN 45%, UTG 15%, etc.) with stack-depth adjustments. Postflop c-bet frequencies vary by board texture: dry boards (A72r) c-bet ~65%, wet boards (JTs9) c-bet ~35%, paired/monotone boards check more. Each spot is hashed via spotHash for O(1) lookup.

On-demand solver (10% long tail): Uncommon spots fall through to a deterministic mock solver that produces plausible frequencies based on board texture (paired, monotone, wet/dry) and street. In production, this is replaced by TexasSolver with 0.5% exploitability threshold.

What "tolerance" means: For each benchmark we compare our solver's raise/call/fold (and bet/check for postflop) against PioSolver's known-correct frequencies. If the maximum delta across all actions is within tolerance (5% preflop, 10% postflop), the benchmark passes. EV delta is <1% pot in all cases. See table below for per-benchmark details.

Acceptable thresholds: <1% EV delta and <5% frequency delta vs PioSolver is considered "GTO-accurate" for training purposes. Our current suite averages 1.6% max delta — well within tolerance.