Solver

Watch an equilibrium converge

Counterfactual Regret Minimization plays itself, accumulates regret at each of the 12 decision points, and steers toward the strategy nobody can beat. The chart below is the only honest scoreboard: how much a perfect adversary could win against the current answer.

Exploitability
chips/hand a perfect counter wins
Iterations
self-play passes
Value to Player 1
exact answer: −0.05556
Bluff rate α
which equilibrium it landed on
No data yet.

Both axes are logarithmic, on purpose. Convergence is a power law, and a power law is a straight line here. The dashed grey line is the theoretical O(1/√T) rate. Vanilla CFR tracks it closely. CFR+ and Linear CFR fall away beneath it — that gap is the entire point of those variants, and it is invisible on linear axes.

Which parts of CFR+ actually matter

All three variants under an identical 20,000-iteration budget.

Performance against rule-based opponents

Exact expected value by enumeration, plus a 100,000-hand simulation with confidence intervals. Every matchup is played from both seats.