{"id":"ed2981ce-0877-40cd-94a1-25cb5fb5a30f","arxiv_id":"2602.15225","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For large n, all pure and symmetric mixed Nash equilibria of a general position-optimization game are extreme and converge to the target-distribution projection at rate O(1/n).","lead":"This paper defines position-optimization games, where each of n players chooses a position to capture the largest possible share of targets drawn from a known distribution, and proves that once n is large enough, all Nash equilibria concentrate on a finite set of 'pseudo-targets' and converge to the underlying target distribution at rate O(1/n). The result unifies and extends prior work in Hotelling location games, forecasting competitions, Voronoi games, and spatial voting,","discovery_kind":"unification","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 6, the basis for symmetric-mixed-equilibrium existence, is false as stated; a compact symmetric constant-sum game can have no symmetric mixed equilibrium. The paper's mixed-existence claim is therefore unproven.","rationale":"The reader's verdict was CONDITIONAL, with the weakest assumption being Condition 1's finiteness and the rationale noting that the Reny existence proof does not properly verify the required conditions. My stress test agrees that the mixed-equilibrium analysis is the main soft spot, but identifies a sharper and more concrete problem: Theorem 6, the general existence lemma, is false as stated. This is not a mere gap in a proof step; the theorem itself has a counterexample. Since the central claim includes existence of symmetric mixed Nash equilibria, the current manuscript's mixed half is unsupported. The pure-equilibrium results and the algorithmic construction may remain correct, and the position-optimization game may still have symmetric mixed equilibria, so I do not reject the paper outright. Rather, the verdict stays CONDITIONAL: the authors must supply a valid existence proof for the specific game (or replace Theorem 6 with a correct sufficient condition), in addition to repairing the other mixed-equilibrium lemmas. I therefore partially agree with the reader: we identify overlapping concerns, but my load-bearing concern is the false general theorem rather than the finiteness of X*.","tokens_in":26951,"tokens_out":38920,"duration_ms":352857,"concrete_test":"Verify the counterexample to Theorem 6. Consider the two-player symmetric zero-sum game on X=[0,1] with u(a,b)=1 if (b=1 and a<1) or (a>b and a,b<1); u(a,b)=-1 if (a=1 and b<1) or (b>a and a,b<1); u(a,a)=0. Prove that no probability σ satisfies u(a,σ)≤0 for all a: if σ({1})>0, Nash requires σ=δ_1, but u(a,δ_1)=1 for a<1; if σ({1})=0, let M=sup supp σ, then u(M,σ)=0 forces σ=δ_M, but u(a,δ_M)=1 for a>M. If this holds, Theorem 6 is false and the mixed-existence proof must be replaced by a game-specific argument.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The paper's only proof of symmetric mixed equilibrium existence is Theorem 6: 'Any symmetric, constant-sum game with pure strategy set corresponding to a compact Hausdorff space has a mixed symmetric equilibrium.' The proof claims payoff security from the fact that payoffs are constant on the diagonal, but Reny's payoff security requires, for every profile and every neighborhood of opponents' strategies, a single unilateral deviation that works. Restricting opponent deviations to the diagonal is not a neighborhood in the product topology and does not establish payoff security.\n\nThe theorem is in fact false. Take X=[0,1] and define u(a,b)=1 if (b=1 and a<1) or (a>b and a,b<1); u(a,b)=-1 if (a=1 and b<1) or (b>a and a,b<1); and u(a,a)=0. This game is antisymmetric, hence constant-sum with c=0, and X is compact Hausdorff. It has no symmetric mixed equilibrium: if σ has mass q on 1, equilibrium requires q=1, but δ_1 is beaten by any a<1; if q=0, let M=sup supp σ, and equilibrium forces all mass at M, but any a>M beats M. Thus Theorem 6 is false, and the existence proof for the mixed half of the central claim collapses. The paper supplies no alternative existence argument for symmetric mixed equilibria in the position-optimization game.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces position-optimization games, in which n symmetric agents choose positions to maximize the expected mass of targets for which they are closest, with ties split evenly. Under Condition 1 (a.s. unique pseudo-target and finite pseudo-target set X*), it proves: for n ≥ 2/p0, an extreme pure Nash equilibrium exists and can be constructed in O(|X*|n) time; for n > 1/p0 all pure equilibria are extreme and cover X*; and the empirical equilibrium distribution converges to the projected target distribution P in KL divergence with the explicit bound log((⌊cn⌋+1)/⌊cn⌋). It further claims existence for all n > 1 of symmetric mixed equilibria (Theorem 4 via the general Theorem 6) and, for large n, extremeness and |σ_x − p_x| ≤ 1/n (Theorem 5). Applications to forecasting competitions, finite and classic Hotelling games, spatial voting, and discrete Voronoi games are then drawn.","tokens_in":27271,"tokens_out":22527,"duration_ms":204714,"significance":"If the theorems were correct, the paper would be a valuable unifying treatment: it gives finite-n existence, structural predictions (all equilibria extreme), and the first O(1/n) convergence rates for a class that includes forecasting competitions and Hotelling variants. The pure-strategy side is constructive, with an explicit algorithm and a tight threshold; these are concrete contributions. However, the mixed-strategy existence claim rests on a false general theorem and an invalid Reny-security proof, so the mixed-strategy half of the advertised contribution is not established. The pseudo-target framework and the explicit rates are genuine strengths, but the mixed-equilibrium results need new proofs before they can be accepted.","major_comments":[{"comment":"Theorem 6 is false as stated, and its proof does not verify Reny payoff security. The proof only checks symmetric diagonal profiles: if every opponent plays σ′, the deviator plays σ′ and secures c/n. Payoff security requires, for every profile and ε > 0, a single deviation that remains ε-optimal against every strategy tuple in a neighborhood of the given opponents' tuple; the diagonal is not a neighborhood in the product topology. The statement is not merely unproved: on X=[0,1], define u(a,b)=1 if (b=1 and a<1) or (a>b and a,b<1); u(a,b)=−1 if (a=1 and b<1) or (b>a and a,b<1); u(a,a)=0. This is a compact symmetric constant-sum game with no symmetric mixed equilibrium. Hence Theorem 4 and the mixed-existence claims in the abstract and Section 5 are unsupported.","section":"Appendix C.1, Theorem 6"},{"comment":"The proof of the upper bound in Theorem 5 is not complete. Lemma 12 claims monotonicity of the function Gbar by arguing that N′>0 and 0<D′<1 imply N/D is increasing; this is not a valid criterion for a quotient. Proposition 4 then applies Gbar^{-1} outside the domain on which monotonicity was stated and contains mismatched references ('Lemma 3' where Observation 3 is meant). Since the conclusion σ_x ≤ p_x + 1/n depends on this step, the upper-bound half of Theorem 5 is not established as written.","section":"Appendix C.2, Lemma 12 / Proposition 4"}],"minor_comments":[{"comment":"The displayed bound after Eq. (4) should be |log(k/(nP))| ≤ log((⌊cn⌋+1)/⌊cn⌋), not the ratio (⌊cn⌋+1)/⌊cn⌋; as printed, the subsequent KL inequality does not follow. This appears to be a local typo.","section":"Theorem 3 proof, after Eq. (4)"},{"comment":"The hypercube vertices x1 and x3 are both labeled {1,0}; presumably one is {0,1}. In the same proof, the text refers to ui(x1,x_−i) when discussing an agent on x2; the notation is confusing and should be cleaned up.","section":"Lemma 4"},{"comment":"The inequality 'P(x)>2/c≥2/n' should presumably be 'P(x)>2c≥2/n'; as written the claimed chain is false. The following display has a similar typo, using c/(⌊cn⌋) instead of the intended 2c/(⌊cn⌋) argument.","section":"Appendix B.4, Lemma 5 proof"},{"comment":"The line 'we have |X*|≤p0' should be |X*|≤1/p0; the subsequent bound uses the corrected inequality. This is a typographical issue, but it makes the displayed derivation hard to follow.","section":"Lemma 11 proof"},{"comment":"The abstract contains the typo 'psuedo-targets'; the term should be 'pseudo-targets'.","section":"Abstract"}],"recommendation":"major_revision","confidential_remarks":"The main risk is the false general existence theorem (Theorem 6). The pure-strategy results are likely salvageable and valuable, but the paper cannot be accepted with the mixed-existence claim resting on an invalid proof. I would ask the authors to either supply a correct existence proof for the specific position-optimization game or remove/downgrade the mixed-existence claims. The upper-bound proof for Theorem 5 also needs a rigorous rewrite. If no correct mixed-existence argument can be provided, the mixed-strategy claims should be presented as conjectures rather than theorems."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The pure-equilibrium half of this paper is worth your time. The position-optimization framework is a nice unifying abstraction, and Theorems 1–3 are genuinely new: explicit extreme equilibria, an O(1/n) KL convergence rate, and a clean constructive algorithm with a runtime bound. I checked the pure proofs with some care and they hold together. The forecasting-competition corollaries, especially the finite-n characterization and the practical interpretation, are interesting and clearly presented.\n\nThe mixed-equilibrium half has a load-bearing flaw. Theorem 6 claims every symmetric constant-sum game on a compact Hausdorff pure set has a symmetric mixed equilibrium. That is false. A simple counterexample: on X=[0,1], define u(a,b)=1 if (b=1 and a<1) or (a>b and a,b<1), u(a,b)=-1 if (a=1 and b<1) or (b>a and a,b<1), and u(a,a)=0. This is antisymmetric (constant-sum c=0) but has no symmetric mixed equilibrium. The proof's error is that it substitutes constancy on the diagonal for Reny's payoff security, but payoff security requires a deviation that works against every profile in a neighborhood of the opponents' strategies; the diagonal is not a neighborhood in the product topology. Since Theorem 4 relies on Theorem 6, the existence of symmetric mixed equilibria in your game is unproven. The upper-bound part of Theorem 5 also rests on Lemma 12, which I found under-specified and not convincing as written; the monotonicity claim there needs a careful redo.\n\nThe pure results alone are a real contribution, and the paper is generally well-written and self-contained. But the central claim of symmetric mixed equilibrium existence, for all n, is currently unsupported. This is a fixable gap rather than a refutation of the entire project, but it is serious: the authors need to either prove existence for their specific position-optimization game (which may be possible using its extra structure) or substantially weaken the claims. The general theorem they state is not true.\n\nWho should read this? Game theorists working on Hotelling models or forecasting competitions, and anyone interested in large finite games. It deserves a serious referee, but with a clear expectation of major revision. I would not cite it in its current form; the mixed results are too shaky. Bring it to reading group if you want a good example of a plausible-sounding theorem that breaks on a compact strategy space.","headline":"Solid pure-equilibrium and convergence-rate results, but the symmetric mixed equilibrium existence theorem is false as stated — that half of the paper needs a real fix.","tokens_in":27757,"tokens_out":2881,"would_cite":false,"duration_ms":35013,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["91A10","91A06","91B72"],"pacs":[],"model":"deepseek-v4-flash","headline":"Once the number of players is large, every Nash equilibrium in a position-optimization game concentrates on a finite set of pseudo-targets, and the distribution of play converges to the projected target distribution at rate O(1/n).","keywords":["position-optimization games","Nash equilibrium","pseudo-targets","Hotelling games","forecasting competitions","convergence rates","symmetric mixed equilibrium","wisdom of crowds"],"falsifier":"Take a game satisfying Condition 1 with n above the mixed-equilibrium threshold, compute a symmetric mixed equilibrium, and check whether any pseudo-target x has |σ_x − p_x| > 1/n; a single violation would refute Theorem 5. More decisively, construct a game with an infinite pseudo-target set (a continuous consumer distribution in Hotelling) and test whether some equilibrium's play distribution fails to converge to P at rate O(1/n).","tokens_in":26838,"feed_emoji":"🎯","tokens_out":7359,"duration_ms":62261,"temperature":0.7,"pith_summary":"This paper establishes that in any position-optimization game—a symmetric game in which n players pick positions to capture the mass of targets to which they are closest—once n is large enough, every Nash equilibrium (pure or symmetric mixed) is extreme: all players play on a finite set of pseudo-targets, the positions that are optimal for some target. Furthermore, the equilibrium distribution over pseudo-targets converges to the projected target distribution P at rate O(1/n). This gives the first convergence rates for Hotelling games and forecasting competitions, and the first general equilibrium characterization for forecasting competitions with finite n. The unifying insight is that large competitions produce a wisdom-of-crowds effect: individually strategic play collectively mimics the underlying distribution.","feed_headline":"Large position games converge to the target distribution at rate 1/n","feed_subtitle":"Unifies Hotelling games and forecasting competitions, bounding how fast play mimics the target distribution.","key_machinery":"The central object is the pseudo-target set X* = {x : x minimizes d(x,y) over X for some y}, together with the projected distribution P(x) = Q({y : x*(y) = x}). For pure equilibria, the argument rests on the observation that a strategy profile covering X* makes all profitable deviations lie inside X*, and on a greedy algorithm (Algorithm 1) that builds an equilibrium by initially placing about 2P(x)/p0 agents on each x and then iteratively adding agents to the position maximizing P(x)/(k+1). For mixed equilibria, the key identity is the utility decomposition into events E1, E2, E3; defining G(σ_x) = (σ_x − σ_x^n)/(1 − σ_x^n − (1−σ_x)^n), equilibrium forces σ_x ≈ G^{-1}(p_x), and a coupon-col","core_discovery":"The paper's central claim is that, under Condition 1 (no ties with positive probability and a finite pseudo-target set X^*), the large-n behavior of the game G_n is fully tractable. For n >= 2/p0, a pure Nash equilibrium exists, is extreme, and has at least two agents on every pseudo-target; for n > 2/p0 this holds for every pure equilibrium. For symmetric mixed equilibria, once n > max{43, 8(4/p0)log(1/p0)}, every such equilibrium is extreme and its per-position probabilities satisfy |σ_x − p_x| ≤ 1/n. In both cases the play distribution converges to P, the projection of the target distribution Q onto X*, with an explicit O(1/n) rate; the paper also proves the pure-existence threshold is ti","pith_inferences":["The authors conjecture in Section 6 that convergence holds without the finiteness condition; if true, the rate likely depends on the geometry of X*, and a natural test is whether continuous Hotelling games also exhibit O(1/n) convergence in total variation.","The mixed-equilibrium result implies a falsifiable prediction for real forecasting contests: winners' submissions should be less calibrated (more extreme) than the average forecaster, and the gap should shrink like 1/n; this could be tested on archived Kaggle-style data.","The wide gap between the pure threshold (2/p0) and the mixed threshold (exponential in 1/p0) suggests that randomizing players need far larger populations to concentrate than deterministic ones; contest designers might exploit this by making participation sizes public.","The componentwise bound |σ_x − p_x| ≤ 1/n gives a simple model-checking test: repeated plays of a symmetric mixed equilibrium should produce empirical frequencies within ~1/n of P for every pseudo-target; systematic deviations indicate misspecification or non-equilibrium behavior."],"forward_implications":["In forecasting competitions with enough forecasters, equilibrium reports are extreme (deterministic outcome predictions) and the empirical distribution of reports converges to the outcome distribution Q, so collecting many reports recovers Q even though each forecaster is strategic.","In Hotelling games with finite or discrete consumer distributions, retailers locate on first-preference locations and the empirical distribution of locations approaches the consumer distribution with an explicit O(1/n) KL-divergence bound, extending prior results to non-differentiable consumer distributions.","The symmetric mixed equilibrium converges to P componentwise within 1/n, meaning aggregate random play is nearly exact even though each player randomizes.","The pure-equilibrium existence threshold n ≥ 2/p0 is tight; below it equilibria may fail, exactly identifying when coordination guarantees a stable outcome.","The convergence-rate bound applies to every equilibrium (not just the constructed one), so the wisdom-of-crowds guarantee is a property of all large equilibria."],"fun_headline_variants":["Position games: pure and mixed equilibria converge at rate 1/n","Large-n position games settle on a finite pseudo-target set","Nash equilibria in position games hit 1/n convergence","Position-optimization games: 1/n convergence to target distribution","Hotelling and forecasting games share 1/n convergence"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The whole proof depends on the set of pseudo-targets being finite and on ties in the closest-position mapping having zero probability; without finiteness, the covering and coupon-collector arguments collapse.","fun_headline_variants_meta":{"raw":{"variants":["Position games: pure and mixed equilibria converge at rate 1/n","Large-n position games settle on a finite pseudo-target set","Nash equilibria in position games hit 1/n convergence","Position-optimization games: 1/n convergence to target distribution","Hotelling and forecasting games share 1/n convergence"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000191,"raw_usage":{"total_tokens":1172,"prompt_tokens":728,"completion_tokens":444,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":472,"completion_tokens_details":{"reasoning_tokens":358}},"tokens_in":472,"tokens_out":444,"duration_ms":4467,"temperature":1.0,"reasoning_tokens":358,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T22:56:18.752725+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a game satisfying Condition 1 with n above the mixed-equilibrium threshold, compute a symmetric mixed equilibrium, and check whether any pseudo-target x has |σ_x − p_x| > 1/n; a single violation would refute Theorem 5. More decisively, construct a game with an infinite pseudo-target set (a continuous consumer distribution in Hotelling) and test whether some equilibrium's play distribution fails to converge to P at rate O(1/n).","supporting_citations":[],"review_version":1}