# yaml-language-server: $schema=https://raw.githubusercontent.com/mathlib-initiative/formalization.yaml/main/schema/formalization.schema.json version: "v0.4" project: name: "Kuhn poker: the value −1/18, all optimal strategies, and the solution at every ante and bet" description: >- A Lean 4 development of Kuhn's simplified two-person poker (Kuhn 1950): three cards 1 < 2 < 3, one dealt to each player with the six deals equiprobable, an ante of a from each player and a single bet of b, player I acting first, the five courses of play of Kuhn's p. 98 with payoffs a (pass-pass, to the higher card), a to II (pass-bet-pass), a + b to the higher card (pass-bet-bet), a to I (bet-pass) and a + b to the higher card (bet-bet). The definitions are built from the game tree: a behaviour strategy of I is a pair of functions on the three cards (the probability of betting first, the probability of calling after passing and facing a bet), one of II likewise (the probability of betting after a pass, the probability of calling a bet); the payoff is the average over the six deals; a strategy guarantees v when it is valid and earns at least v (holds the opponent to at most v) against every valid strategy of the opponent; the game has value v when each player has a strategy guaranteeing v; an equilibrium is a saddle point. For Kuhn's game a = b = 1, kernel-checked: the value is -1/18 (a game has at most one value), an equilibrium exists, every equilibrium pays I exactly -1/18, and a pair is an equilibrium iff each strategy guarantees -1/18; II's optimal strategy is unique at all six information sets (bet after a pass with the 1, 2, 3 with probabilities 1/3, 0, 1; call with probabilities 0, 1/3, 1 - Kuhn's (xi, eta) = (1/3, 1/3) with the four coordinates he removed as superfluous now proved forced); I's optimal strategies are exactly Kuhn's one-parameter family (bet the 1 with probability alpha and fold it, never bet the 2 first and call with it with probability alpha + 1/3, bet the 3 with probability 3 alpha and call with it, for alpha in [0, 1/3], the call with the 3 being unconstrained exactly when 3 alpha = 1, where that information set is unreached); in the normal form with Kuhn's coding of the 27 and 64 pure strategies, his essential 8 x 4 matrix is six times the game matrix, his twelve basic optimal mixed strategies (A)-(L) and II's two guarantee -1/18, and a mixed strategy is optimal iff its realization plan lies on the family (I) or its marginals are II's strategy (II). For every ante a > 0 and bet b > 0 (Kuhn's pp. 102-103, stated there without proof, and the region b > 2a he does not state): the value is -b^2/(6(2a + b)) for b <= a, -b(2a - b)/(6(2a + b)) for a <= b <= 2a and 0 for b >= 2a, with explicit guaranteeing strategies in each case; in Case 3 the calling probability eta with the 2 ranges over [(2a - b)/(2a + b), b/(2a + b)], and the interval printed on p. 102, b/(2a + b) <= eta <= (a + b)/(2a + b), fails at its upper end (at a = 2, b = 3 and eta = 5/7 II's Case 3 strategy does not guarantee the value). Every guarantee is a finite nonnegative-product identity (the gap between the payoff and the value is a sum of products of the opponent's probabilities and their complements with coefficients nonnegative on the case's region: polynomials in a, b and the free parameter, at general stakes built from a, b, 2a + b, a - b, 2a - b, 4a - b, b - 2a and the interval factors (2a + b)eta - (2a - b), b - (2a + b)eta, 2(a - b) - (2a + b)xi, and, in the optimal-set theorems, the slacks of their stated inequalities), and every "only if" is a linear certificate from finitely many pure strategies (the coefficients rationals at a = b = 1 and, at general stakes, polynomials in a, b and the free parameter nonnegative on the case's region); no formal library found by the dated search holds the game. Beyond Kuhn's statements: the two certificate identities are stated as theorems; the equilibria of Kuhn's game are exactly the pairs (kuhnI α k, kuhnII) with α in [0, 1/3], k in [0, 1] and 3α = 1 or k = 1; and at general stakes the optimal sets are determined - I's strategy is unique for b < a (up to the unreached call with the 3) and for a < b <= 2a, is Kuhn's family on the diagonal b = a at every scale, and for b >= 2a never bets the 1 or the 2 first, folds them, and with the 3 is cut out by one inequality (a single strategy at b = 2a; beyond it a two-dimensional set in which I may also bet the 3 first, with probability at most (b - 2a)/b), so Kuhn's one-parameter family is optimal only at b = a; II's strategy is unique on the diagonal at every scale and a segment for b <= a (the bet with the 2 after a pass free in [0, 2(a - b)/(2a + b)], so Kuhn's 'unique optimal mode of behavior' in Case 1 holds in his behaviour parameters only); for a <= b < 2a its bets after a pass are forced, (b/(2a + b), 0, 1), and its calls form a polytope of three inequalities; for b > 2a it never bets the 1 or the 2 after a pass, never calls with them, and calls with the 3 with probability at least 4a/(2a + b), its bet with the 3 free; and on the line b = 2a its optimal set is a five-dimensional polytope (the bet with the 2 after a pass zero and, with bets (p, 0, r) and calls (s, t, w), 2p <= r, s + t <= p, 1 <= t + w, 1 + p + s <= 2w; nine vertices, replayed); at every ante a > 0 and bet b > 0 every equilibrium pays the value and a pair is an equilibrium iff each strategy guarantees it, so the equilibria are exactly the pairs of optimal strategies, on b = 2a explicitly; the corrected Case 3 interval is sharp, and the printed one holds only at its lower end. A second correction, a misprint, is noted in the documentation: p. 101's table gives β = 5/12 for (I) and (J), where p. 100's mixtures and p. 101's own formula give 7/12. authors: ["JD Jones"] responsible_maintainers: ["JD Jones"] license: "MIT" classification: arxiv: ["cs.GT"] msc2020: ["91A05", "91A18", "91A60"] sources: - title: "A simplified two-person poker" authors: ["H. W. Kuhn"] id: "doi:10.1515/9781400881727-010" type: "paper" location: "In: H. W. Kuhn, A. W. Tucker (eds.), Contributions to the Theory of Games, Volume I, Annals of Mathematics Studies 24, Princeton University Press (1950), 97-103 (Crossref dates the DOI 1951-12-31 and gives pp. 97-104, the scan's eighth page being blank; the catalogues and zbMATH give 1950, 97-103); the whole paper read: p. 97 (the rules), p. 98 (the five courses of play and the coding of pure strategies), p. 99 (the essential 8 x 4 matrix), p. 100 (the twelve and two basic optimal strategies, the value -1/18), p. 101 (the behaviour parameters and the one-parameter family), pp. 102-103 (the general ante a and bet b, Cases 1-4, stated without proof)" relationship: "formalizes" author_endorsement: "not-contacted" note: >- The source of the game and of every classical statement: the rules and payoffs (p. 97-98), the essential matrix (p. 99), the basic optimal strategies and the value (p. 100), the behaviour-parameter family (p. 101), II's single solution (p. 102), and the general-stakes solutions with their values (pp. 102-103). The proofs here are not Kuhn's: his completeness argument goes through the superfluousness of dominated strategies and the kernel theory of discrete games (the papers of this study he cites), whereas every inequality here is a finite nonnegative-product identity and every "only if" a linear certificate from finitely many pure strategies, with the coordinates Kuhn removes as superfluous (three of I's, four of II's) proved forced rather than cited (I's call with the 3 except where 3 alpha = 1 leaves it unreached). Two corrections to the printed text and one scope note: the Case 3 interval for eta (case_three_II_iff, case_three_printed_interval, case_three_printed_upper_end_fails); p. 101's beta = 5/12 for (I) and (J), where 7/12 is meant (noted in the documentation, not a theorem); and the scope of "both players have a unique optimal mode of behavior" in Case 1, true in his behaviour parameters, while in the full game II may also bet the 2 after a pass with any probability in [0, 2(a - b)/(2a + b)] (case_one_II_iff). - title: "The Mathematics of Poker" authors: ["Bill Chen", "Jerrod Ankenman"] id: "isbn:9781886070257" type: "book" location: "ConJelCo, Pittsburgh (2006), x + 382 pp.; not read" relationship: "background" author_endorsement: "not-contacted" note: "Named for context as a later poker-theory exposition that treats three-card games with a pot parameter; not read, nothing is taken from it, and no statement of this entry is attributed to it." - title: "Basic solutions of discrete games" authors: ["L. S. Shapley", "R. N. Snow"] id: "doi:10.1515/9781400881727-004" type: "paper" location: "In: H. W. Kuhn, A. W. Tucker (eds.), Contributions to the Theory of Games, Volume I, Annals of Mathematics Studies 24, Princeton University Press (1950), 27-35; not read" relationship: "background" author_endorsement: "not-contacted" note: >- The kernel theory Kuhn's completeness proof invokes on pp. 100-101 ("all optimal strategies are convex linear combinations of the strategies given above"). Nothing from it is formalized or relied on: the optimal sets are characterized here by linear certificates, and Kuhn's convex-hull statement is replayed by scripts/check_kuhn.py (an exact enumeration finds exactly his twelve and two vertices) but is not stated as a Lean theorem, whose "exactly" clause is the realization-plan form. - title: "Solutions of finite two-person games" authors: ["D. Gale", "S. Sherman"] id: "doi:10.1515/9781400881727-005" type: "paper" location: "In: H. W. Kuhn, A. W. Tucker (eds.), Contributions to the Theory of Games, Volume I, Annals of Mathematics Studies 24, Princeton University Press (1950), 37-49; not read" relationship: "background" author_endorsement: "not-contacted" note: "Kuhn's p. 100 takes the terms essential and superfluous from it. Nothing from it is used here." - title: "Solutions of discrete, two-person games" authors: ["H. F. Bohnenblust", "S. Karlin", "L. S. Shapley"] id: "doi:10.1515/9781400881727-006" type: "paper" location: "In: H. W. Kuhn, A. W. Tucker (eds.), Contributions to the Theory of Games, Volume I, Annals of Mathematics Studies 24, Princeton University Press (1950), 51-72; not read" relationship: "background" author_endorsement: "not-contacted" note: "Kuhn's p. 100 counts the linearly independent optimal strategies (6 and 2) by its results. Nothing from it is used here." - title: "Efficient computation of behavior strategies" authors: ["B. von Stengel"] id: "doi:10.1006/game.1996.0050" type: "paper" location: "Games and Economic Behavior 14 (1996), no. 2, 220-246; not read" relationship: "background" author_endorsement: "not-contacted" note: "The sequence form and realization plans, the terms of the normal-form statements and of the affine bridge; the bridge is proved here for Kuhn's game only, nothing is imported." - title: "Zur Theorie der Gesellschaftsspiele" authors: ["J. von Neumann"] id: "doi:10.1007/BF01448847" type: "paper" location: "Mathematische Annalen 100 (1928), 295-320; not read" relationship: "background" author_endorsement: "n/a" note: "The minimax theorem named in the note on PALOMAR-2026-10-01-000004; not used here, where the value is exhibited by certificates." - title: "On general minimax theorems" authors: ["M. Sion"] id: "doi:10.2140/pjm.1958.8.171" type: "paper" location: "Pacific Journal of Mathematics 8 (1958), no. 1, 171-176; not read" relationship: "background" author_endorsement: "n/a" note: "Named in the note on PALOMAR-2026-10-01-000004 (its proof route in Mathlib); not used here." - title: "Extensive games and the problem of information" authors: ["H. W. Kuhn"] id: "doi:10.1515/9781400881970-012" type: "paper" location: "In: H. W. Kuhn, A. W. Tucker (eds.), Contributions to the Theory of Games, Volume II, Annals of Mathematics Studies 28, Princeton University Press (1953), 193-216; not read" relationship: "background" author_endorsement: "not-contacted" note: "Behaviour strategies and perfect recall, named in the note on elazarg/GameTheory; the passage between mixed and behaviour strategies here is proved directly for this game." - title: "The Lean 4 Theorem Prover and Programming Language" authors: ["Leonardo de Moura", "Sebastian Ullrich"] id: "doi:10.1007/978-3-030-79876-5_37" type: "paper" location: "Automated Deduction - CADE 28, Lecture Notes in Computer Science, Springer, 2021, 625-635" relationship: "background" author_endorsement: "n/a" note: "The proof assistant (leanprover/lean4:v4.35.0-rc2). Nothing mathematical is taken from it." - title: "The Lean mathematical library" authors: ["The mathlib Community"] id: "doi:10.1145/3372885.3373824" type: "paper" location: "Proceedings of the 9th ACM SIGPLAN International Conference on Certified Programs and Proofs (CPP 2020), 367-381" relationship: "background" author_endorsement: "n/a" note: "The library the development imports (tag v4.35.0-rc2): real arithmetic, Fin, finite sums, matrices and the tactics linarith, linear_combination, positivity and norm_num." related_formalizations: - id: "PALOMAR-2026-10-01-000004" relationship: "other" note: >- minimax-theorem-lean (Arthur742Ramos, registered 2026-10-01): for every finite two-player zero-sum game, the sup-inf and inf-sup of the expected payoff over mixed strategies agree (von Neumann's minimax theorem; its three definitions are comparator holes, which the repository defines so, proving a saddle point through Mathlib's Sion theorem). Its theorem implies that Kuhn poker's 27 x 64 normal form has a value; it does not compute it. This entry exhibits the value and both optimal sets directly, by certificates, in behaviour and in mixed strategies, and shares no code with it. - id: "PALOMAR-2026-09-08-000002" relationship: "other" note: >- Verified regret matching and finite-time equilibrium computation (Arthur742Ramos, registered 2026-09-08): regret bounds and approximate equilibria of finite zero-sum matrix games by self-play; its repository metadata states that no extensive-form or poker performance theorem is claimed. Kuhn poker is a standard test game in that literature; no equilibrium of it is stated there. No code is shared. - id: "https://github.com/elazarg/GameTheory/tree/e33b1317302b3413b2d690862eabbbe067a7d499" relationship: "other" note: >- A Lean 4 game-theory library (extensive-form games, behaviour strategies and perfect recall after Kuhn 1953, counterfactual-regret modules; main at e33b131, 2026-10-09, read by tree path and content search, and re-read 2026-10-10 with no poker path on main or on the branch pushed that day). It holds no poker instance and no value of Kuhn poker. No code is shared; this entry builds its game from the tree without a library. status: scope: >- Thirty-seven compared theorems over thirty-one Challenge definitions, all built from Kuhn's game tree. For Kuhn's game (a = b = 1): the value -1/18 in both guarantee and equilibrium form, uniqueness of the value, the two certificate identities, the explicit equilibrium set, II's unique optimal strategy, I's optimal set as Kuhn's family with the unreached-information-set clause, Kuhn's essential matrix, his twelve and two basic optimal mixed strategies, the value of the normal form, and the exact optimal sets of the normal form in realization-plan and marginal form. For every ante and bet: the value function and guaranteeing strategies (a, b > 0) in the three regions b <= a, a <= b <= 2a, b >= 2a, with the corrected Case 3 interval for eta, and both players' optimal sets in every region, the diagonal and the line b = 2a included (on that line II's set is a five-dimensional polytope cut out, within the valid strategies, by one equation and four inequalities, the six-product identity behind it a theorem); every equilibrium pays the value and a pair is an equilibrium iff each strategy guarantees it, at every ante and bet, the equilibria on b = 2a explicit. Not claimed as compared theorems: the nine vertices and the dimension of II's set on b = 2a (replayed by scripts/check_boundary.py); Kuhn's convex-hull form of the optimal polytopes (its vertices replayed by scripts/check_kuhn.py, not stated; equivalent on paper to the realization-plan form stated); anything about more than three cards. The intentional placeholders of Challenge.lean are excluded from proof counts; Solution.lean does not import Challenge.lean. sorry_count: 0 sorry_in_definitions: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] main_results: - declaration: "KuhnPoker.hasValue_one_one" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.hasValue_unique" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.exists_isEquilibrium" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.payoff_of_isEquilibrium" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.isEquilibrium_iff" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.guarantees₂_iff" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.guarantees₁_iff" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.essentialMatrix_eq" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.basic₁_guarantees" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.basic₂_guarantees" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.hasMixedValue" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.guaranteesMixed₁_iff" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.guaranteesMixed₂_iff" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.hasValue_general" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.kuhnValue_one_one" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.guarantees_case_one" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.guarantees_case_three" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.guarantees_case_four" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.upper_certificate" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.lower_certificate" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.isEquilibrium_iff_explicit" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.case_one_I_unique" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.case_one_II_iff" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.diagonal_II_unique" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.diagonal_I_iff" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.case_three_I_unique" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.case_three_II_iff" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.case_three_II_set" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.case_three_printed_interval" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.case_three_printed_upper_end_fails" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.case_four_I_iff" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.case_four_II_iff" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.boundary_II_certificate" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.boundary_II_iff" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.payoff_of_isEquilibrium_general" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.isEquilibrium_general_iff" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" - declaration: "KuhnPoker.boundary_isEquilibrium_iff" file: "Solution.lean" sorry_count: 0 axioms: ["propext", "Classical.choice", "Quot.sound"] comparator_config: "comparator.json" automation: methods: - method: "agent" models: ["Claude Opus 5.5 (Anthropic)", "Claude Fable 5.1 (Anthropic)"] framework: "Claude Code with parallel subagents" tool_setup: >- The project was directed by its author. Claude (Anthropic) in Claude Code was used throughout: the reading of Kuhn's paper, the statements and their check against it, the exact certificates, the Lean proofs, the standard-library replay, the prior-art searches and the repository text; Claude Fable 5.1 pinned the statements, wrote the replay checker, merged the proof files and wrote the documents, and Claude Opus 5.5 subagents computed and verified the certificates (exact rational arithmetic, with linear-programming duals as finders), wrote the Lean proofs and audited them. Lean v4.35.0-rc2 and Mathlib v4.35.0-rc2 (commit 065356127b1dc0016f66b7283ce0ce2c4055aa55) are pinned by the committed manifest and lake update is never run; there are no GitHub Actions workflows. There is no native_decide and no decide +kernel: the proofs unfold the payoff to a polynomial over the reals in the twelve probabilities (and a, b), close each inequality by linarith with the certificate's nonnegative products supplied as mul_nonneg hints (the identities by ring, one by linear_combination; no nlinarith), each "only if" by linarith from finitely many pure strategies, and the normal-form statements by an affine bridge from mixed strategies to realization plans and marginals. - method: "agent" models: ["GPT-6 (OpenAI)"] framework: "ChatGPT" tool_setup: >- A pre-submission review of the tree at a46246e (2026-10-10). Its proposals, the five statements on the line b = 2a and the all-stakes equilibria with their proofs and the source guard's regex fix, were checked here with the full verification gate before adoption; the documentation is this repository's. spend_usd: "not tracked" fidelity: divergences: >- Cards are 0 < 1 < 2 for Kuhn's 1, 2, 3; a behaviour strategy is a pair of functions Fin 3 -> ℝ (the Challenge defines Strategy₁ and Strategy₂ as that product type), with validity the condition that every coordinate lies in [0, 1]; the payoff is the sum over the six deals of distinct cards divided by six, with the five courses as a polynomial in the probabilities. Kuhn's behaviour parameters (alpha, beta, gamma; xi, eta) on p. 101 live in the reduced game after dominated strategies are removed; here the strategies carry all twelve coordinates, and the removed ones are proved to take their forced values in every optimal strategy of Kuhn's game (I: fold the 1 to a bet, never bet the 2 first, call with the 3 unless 3 alpha = 1; II: never bet the 2 after a pass, always bet the 3 after a pass, never call with the 1, always call with the 3); on the diagonal b = a the same holds at every scale, and off it some are not forced, as the optimal-set theorems state: I's call with the 3 for b < a (where the 3 is always bet) and for b > 2a; II's bet with the 2 after a pass for b < a, its call with the 1 for a < b <= 2a, its bet with the 3 after a pass for b >= 2a, and its call with the 3 for b > a. The family is indexed by alpha = gamma/3 in [0, 1/3] rather than Kuhn's gamma in [0, 1]. The clause "3 alpha = 1 or k = 1" in guarantees₁_iff is the unreached-information-set convention made explicit: Kuhn's family says nothing about the call with the 3 when it always bets. Optimality is defined as guaranteeing the value against every valid strategy of the opponent (the maximin/minimax form), and shown equivalent to being part of an equilibrium. The normal-form statements use Kuhn's coding of pure strategies verbatim (x_c in {0, 1, 2}; y_c in {0, 1, 2, 3} in binary) and state the optimal sets by realization plans and marginals, not as convex hulls of his vertices. The general-stakes theorems state the value function kuhnValue a b with three regions and closed boundaries (Kuhn's Case 2, b = a, is the common boundary of his Cases 1 and 3; his Case 4 is the single stake b = 2a, extended here to all b >= 2a); the Case 3 witness for II carries the corrected interval for eta, [(2a - b)/(2a + b), b/(2a + b)], in place of the printed b/(2a + b) <= eta <= (a + b)/(2a + b), which the standard-library checker refutes at a = 2, b = 3. review: status: "self-assessed" notes: >- No independent human expert review and no source-author endorsement is claimed; AI agents computed the certificates, wrote the proofs and audited them in separate passes, the statements were checked against Kuhn's paper page by page, and VERIFICATION.md states what is checked and what is not. Mechanical correctness and agreement of formal statements do not establish priority or significance. Kuhn's 1950 paper states the value and the optimal strategies for a = b = 1 and, without proof, the general-stakes solutions; the full-game optimal sets at general stakes, the region b > 2a and the corrections are not in it. His p. 103 remark at b = 2a, 'The strategy for I is unique while II can vary his strategy considerably', names the freedom that boundary_II_iff makes exact. No earlier kernel check of these results was found by the dated search (Palomar's feed and day index, Hexagon, formal-conjectures, openai/math, the Lean game theory libraries, the arXiv API, GitHub code search in Lean, Isabelle, Coq and Agda, 2026-10-09 and 2026-10-10); no first is claimed beyond it. alignment: namespace: "KuhnPoker" statements: - source: "Kuhn 1950, p. 100 (the value -1/18)" lean: "KuhnPoker.hasValue_one_one" module: "Solution" status: "proved" note: "Kuhn's game has value -1/18: each player has a strategy guaranteeing it." - source: "Standard (a zero-sum game has at most one value)" lean: "KuhnPoker.hasValue_unique" module: "Solution" status: "proved" note: "If each player can guarantee v and each can guarantee w, then v = w, at every ante and bet." - source: "Kuhn 1950, p. 100 (the optimal strategies and the value -1/18)" lean: "KuhnPoker.exists_isEquilibrium" module: "Solution" status: "proved" note: "An equilibrium exists." - source: "Kuhn 1950, p. 100" lean: "KuhnPoker.payoff_of_isEquilibrium" module: "Solution" status: "proved" note: "Every equilibrium pays I exactly -1/18." - source: "Standard (exchangeability of optimal strategies in zero-sum games)" lean: "KuhnPoker.isEquilibrium_iff" module: "Solution" status: "proved" note: "A pair is an equilibrium iff each strategy guarantees -1/18." - source: "Kuhn 1950, p. 102 ((xi, eta) = (1/3, 1/3)) and pp. 98-100 (the superfluous strategies)" lean: "KuhnPoker.guarantees₂_iff" module: "Solution" status: "proved" note: "II's optimal strategy is unique: bet after a pass with probabilities 1/3, 0, 1 and call with 0, 1/3, 1, the four forced coordinates proved." - source: "Kuhn 1950, p. 101 (alpha = gamma/3, beta = gamma/3 + 1/3, 0 <= gamma <= 1) and pp. 98-100" lean: "KuhnPoker.guarantees₁_iff" module: "Solution" status: "proved" note: "I's optimal strategies are exactly kuhnI alpha k with alpha in [0, 1/3], k in [0, 1] and (3 alpha = 1 or k = 1)." - source: "Kuhn 1950, p. 99 (the essential matrix, six times the game matrix) with the coding of p. 98" lean: "KuhnPoker.essentialMatrix_eq" module: "Solution" status: "proved" note: "The thirty-two entries, row by row." - source: "Kuhn 1950, p. 100, (A)-(L)" lean: "KuhnPoker.basic₁_guarantees" module: "Solution" status: "proved" note: "Each of the twelve basic optimal mixed strategies guarantees -1/18 in the normal form." - source: "Kuhn 1950, p. 100, the two strategies for II" lean: "KuhnPoker.basic₂_guarantees" module: "Solution" status: "proved" note: "Each of II's two basic optimal mixed strategies guarantees -1/18 in the normal form." - source: "Kuhn 1950, p. 100" lean: "KuhnPoker.hasMixedValue" module: "Solution" status: "proved" note: "The 27 x 64 normal form has value -1/18." - source: "Kuhn 1950, pp. 100-101 (all optimal strategies are convex combinations of (A)-(L)), in realization-plan form" lean: "KuhnPoker.guaranteesMixed₁_iff" module: "Solution" status: "proved" note: "A mixed strategy of I is optimal iff it is a probability vector whose realization plan is (alpha, 0, 3 alpha; 0, alpha + 1/3, 1 - 3 alpha) for some alpha in [0, 1/3]." - source: "Kuhn 1950, pp. 100-101 (II's two basic strategies), in marginal form" lean: "KuhnPoker.guaranteesMixed₂_iff" module: "Solution" status: "proved" note: "A mixed strategy of II is optimal iff it is a probability vector with marginals (1/3, 0, 1; 0, 1/3, 1)." - source: "Kuhn 1950, pp. 102-103, Cases 1-4 (the values), extended to b > 2a" lean: "KuhnPoker.hasValue_general" module: "Solution" status: "proved" note: "For a, b > 0 the game has value kuhnValue a b: -b^2/(6(2a + b)) for b <= a, -b(2a - b)/(6(2a + b)) for a <= b <= 2a, 0 for b >= 2a." - source: "Kuhn 1950, p. 102, Case 2" lean: "KuhnPoker.kuhnValue_one_one" module: "Solution" status: "proved" note: "kuhnValue 1 1 = -1/18." - source: "Kuhn 1950, p. 102, Case 1" lean: "KuhnPoker.guarantees_case_one" module: "Solution" status: "proved" note: "For a, b > 0 and b <= a, Kuhn's Case 1 strategies guarantee kuhnValue a b: I's (b/(2a+b), 0, 1; 0, 2a/(2a+b), 1) and II's (b/(2a+b), 0, 1; 0, (2a-b)/(2a+b), 1)." - source: "Kuhn 1950, p. 102, Case 3, the interval for eta corrected" lean: "KuhnPoker.guarantees_case_three" module: "Solution" status: "proved" note: "For a, b > 0 and a <= b <= 2a, Kuhn's Case 3 strategies guarantee kuhnValue a b: I's (0, 0, 0; 0, (2a-b)/(2a+b), 1) and II's (b/(2a+b), 0, 1; 0, eta, 1) for every eta in the corrected interval [(2a - b)/(2a + b), b/(2a + b)]." - source: "Kuhn 1950, p. 103, Case 4 (b = 2a), and the region b > 2a" lean: "KuhnPoker.guarantees_case_four" module: "Solution" status: "proved" note: "For a, b > 0 and b >= 2a, I's strategy (0, 0, 0; 0, 0, 1) and II's (0, 0, 1; 0, 0, 1) both guarantee 0: I never bets first and calls only with the 3; II bets after a pass only with the 3 and calls only with the 3." - source: "This development (the certificate behind Kuhn's p. 100 value, upper side)" lean: "KuhnPoker.upper_certificate" module: "Solution" status: "proved" note: "For every strategy σ of I, -1/18 - payoff σ kuhnII = (1 - σ.1 0)σ.2 0/6 + σ.1 1/18 + (1 - σ.1 2)(1 - σ.2 2)/6." - source: "This development (the certificate behind Kuhn's p. 101 family, lower side)" lean: "KuhnPoker.lower_certificate" module: "Solution" status: "proved" note: "For every α, k and every strategy τ of II, payoff (kuhnI α k) τ + 1/18 = α/2 τ.2 0 + α/2 (1 - τ.2 2) + (1/3 - α)/2 τ.1 1 + (1/18 + α/6)(1 - τ.1 2) - (1 - 3α)(1 - k)(τ.1 0 + τ.1 1)/2." - source: "Kuhn 1950, pp. 100-102 (the optimal strategies of both players), as the set of equilibria" lean: "KuhnPoker.isEquilibrium_iff_explicit" module: "Solution" status: "proved" note: "IsEquilibrium 1 1 σ τ iff σ = kuhnI α k for some α in [0, 1/3], k in [0, 1] with 3α = 1 or k = 1, and τ = kuhnII." - source: "This development (Kuhn 1950, p. 102, Case 1, states one strategy)" lean: "KuhnPoker.case_one_I_unique" module: "Solution" status: "proved" note: "For a, b > 0 and b < a, I's strategy guarantees the value iff it is (b/(2a+b), 0, 1; 0, 2a/(2a+b), k) for some k in [0, 1]." - source: "This development (Kuhn 1950, p. 102, Case 1: 'both players have a unique optimal mode of behavior' holds for the reduced parameters only)" lean: "KuhnPoker.case_one_II_iff" module: "Solution" status: "proved" note: "For a, b > 0 and b <= a, II's strategy guarantees the value iff it is (b/(2a+b), ξ, 1; 0, (2a-b)/(2a+b), 1) for some ξ in [0, 2(a-b)/(2a+b)]." - source: "Kuhn 1950, p. 102 (Case 2 at every scale)" lean: "KuhnPoker.diagonal_II_unique" module: "Solution" status: "proved" note: "For a > 0, II's strategy guarantees kuhnValue a a iff it is (1/3, 0, 1; 0, 1/3, 1)." - source: "Kuhn 1950, p. 101 (the family), at every scale by homogeneity" lean: "KuhnPoker.diagonal_I_iff" module: "Solution" status: "proved" note: "For a > 0, I's strategy guarantees kuhnValue a a iff it is kuhnI alpha k with alpha in [0, 1/3], k in [0, 1] and (3 alpha = 1 or k = 1)." - source: "This development (Kuhn 1950, p. 102, Case 3, states one strategy)" lean: "KuhnPoker.case_three_I_unique" module: "Solution" status: "proved" note: "For a < b <= 2a, I's strategy guarantees the value iff it is (0, 0, 0; 0, (2a-b)/(2a+b), 1)." - source: "Kuhn 1950, p. 102, Case 3, the interval for η corrected and shown sharp" lean: "KuhnPoker.case_three_II_iff" module: "Solution" status: "proved" note: "For a, b > 0 and a <= b <= 2a, with the other coordinates as pinned, (b/(2a+b), 0, 1; 0, η, 1) guarantees the value iff (2a-b)/(2a+b) <= η <= b/(2a+b)." - source: "This development (Kuhn 1950, p. 102, Case 3, states the η family only)" lean: "KuhnPoker.case_three_II_set" module: "Solution" status: "proved" note: "For a <= b < 2a, II's strategy guarantees the value iff it is valid, its bets after a pass are (b/(2a+b), 0, 1), and its calls satisfy the three stated inequalities." - source: "Kuhn 1950, p. 102, Case 3, the printed interval b/(2a+b) <= η <= (a+b)/(2a+b)" lean: "KuhnPoker.case_three_printed_interval" module: "Solution" status: "proved" note: "For a, b > 0 and a <= b <= 2a, with η on the printed interval b/(2a+b) <= η <= (a+b)/(2a+b), II's strategy (b/(2a+b), 0, 1; 0, η, 1) guarantees the value iff η = b/(2a+b)." - source: "Kuhn 1950, p. 102, Case 3, the printed upper end" lean: "KuhnPoker.case_three_printed_upper_end_fails" module: "Solution" status: "proved" note: "At a = 2, b = 3, η = 5/7 does not guarantee the value." - source: "This development (Kuhn 1950, p. 103, Case 4, states one strategy at b = 2a)" lean: "KuhnPoker.case_four_I_iff" module: "Solution" status: "proved" note: "For a, b > 0 and 2a <= b, I's strategy guarantees 0 iff it is valid, never bets the 1 or the 2 first, folds them, and 4a <= 2a σ.1 2 + (2a+b)(1 - σ.1 2) σ.2 2; at b = 2a this forces σ.1 2 = 0, and for b > 2a the 3 may be bet first with probability at most 1 - 2a/b." - source: "This development (the region b > 2a, which Kuhn does not state; his p. 103 remark on II's freedom at b = 2a is boundary_II_iff)" lean: "KuhnPoker.case_four_II_iff" module: "Solution" status: "proved" note: "For a, b > 0 and 2a < b, II's strategy guarantees 0 iff it is valid, never bets the 1 or 2 after a pass, never calls with them, and 4a <= (2a+b) τ.2 2." - source: "This development (the identity behind II's set on b = 2a)" lean: "KuhnPoker.boundary_II_certificate" module: "Solution" status: "proved" note: "For every a and every pair of strategies with II's bet with the 2 after a pass zero, -3 payoff a (2a) σ τ = a times six products: 2 σ.1 0 (τ.2 1 + τ.2 2 - 1), (1 - σ.1 0) σ.2 0 τ.1 2, σ.1 1 (2 τ.2 2 - 1 - τ.1 0 - τ.2 0), (1 - σ.1 1) σ.2 1 (τ.1 2 - 2 τ.1 0), σ.1 2 (τ.1 0 - τ.2 0 - τ.2 1), 2 (1 - σ.1 2)(1 - σ.2 2) τ.1 0." - source: "This development (Kuhn 1950, p. 103, Case 4, states one strategy for II at b = 2a and remarks on its freedom)" lean: "KuhnPoker.boundary_II_iff" module: "Solution" status: "proved" note: "For a > 0, II's strategy guarantees 0 at b = 2a iff it is valid, never bets the 2 after a pass, and 2 τ.1 0 <= τ.1 2, τ.2 0 + τ.2 1 <= τ.1 0, 1 <= τ.2 1 + τ.2 2, 1 + τ.1 0 + τ.2 0 <= 2 τ.2 2 (a five-dimensional polytope with nine vertices, replayed)." - source: "Standard (the payoff of an equilibrium is the value), at every ante and bet" lean: "KuhnPoker.payoff_of_isEquilibrium_general" module: "Solution" status: "proved" note: "For a, b > 0, every equilibrium pays I exactly kuhnValue a b (a sandwich between the two guaranteeing strategies of hasValue_general)." - source: "Standard (exchangeability of optimal strategies in zero-sum games), at every ante and bet" lean: "KuhnPoker.isEquilibrium_general_iff" module: "Solution" status: "proved" note: "For a, b > 0, a pair is an equilibrium iff each strategy guarantees kuhnValue a b; with the optimal sets this makes the equilibria explicit in every region." - source: "This development (Kuhn 1950, p. 103, Case 4: I's strategy at b = 2a and II's freedom)" lean: "KuhnPoker.boundary_isEquilibrium_iff" module: "Solution" status: "proved" note: "For a > 0, (σ, τ) is an equilibrium at b = 2a iff σ = (0, 0, 0; 0, 0, 1) and τ lies in the polytope of boundary_II_iff."