Kuhn poker: the value −1/18, all optimal strategies, and the solution at every ante and bet

JD Jones

For Kuhn’s three-card poker with positive ante a and a≤b≤2aa \le b \le 2a, corrects the Case 3 calling-probability interval for the middle card to [(2a−b)/(2a+b),b/(2a+b)][(2a-b)/(2a+b), b/(2a+b)] for the specified player-II strategy family. The printed upper endpoint fails, including at a=2a=2 and b=3b=3.

What was already known

Kuhn’s original 1950 paper, p. 102, Case 3. The classical a=b=1a=b=1 game value and strategy work predate this package.

AI involvement AI-centered

JD Jones directed the project. Claude Opus 5.5 subagents computed the rational nonnegative-product certificates and linear certificates from pure strategies, and wrote their Lean proofs. The source does not separately identify who first noticed the printed error.

Importance Modest

The proof corrects one printed interval and gives exact certificates for the corrected endpoints.

Technical verification Formal (Lean)

Pinned source reports Lean verification of case_three_II_iff, case_three_printed_interval and case_three_printed_upper_end_fails with standard axioms.

Formal checking

Pinned source reports Lean verification of case_three_II_iff, case_three_printed_interval and case_three_printed_upper_end_fails with standard axioms.

Checker
JD Jones’s project via Lean 4
Scope
case_three_II_iff, case_three_printed_interval and case_three_printed_upper_end_fails.
Basis
Reported by the linked checker

Human understanding Few

A human explanation of the central model-produced argument is undocumented.

Correction and publication history