{"id":"196fc6c7-3a55-4f94-9ea5-fc63459d1d54","arxiv_id":"1908.08322","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"In a bottleneck queue where customers disagree about service speed, Nash equilibrium arrival patterns separate pessimistic and optimistic customers into mostly disjoint time intervals; a fluid model gives explicit formulas.","lead":"This paper studies how customers who disagree about service speed decide when to arrive at a bottleneck queue. It finds that pessimistic and optimistic customers typically arrive at separate time windows, and that naive learners can match fully rational predictions.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Section 6.2's 'FR' equilibrium is computed by splicing two separate Alg.2 runs with different rate vectors; this does not solve the coupled two-type game, so the ABM/FR closeness claim is unsupported.","rationale":"The reader's concern about Proposition 1 is real but largely a proof gap: the expected-waiting functions in Lemma 4 are continuous in the arrival probabilities, so a standard Kakutani argument for finite anonymous games can likely supply the missing upper hemi-continuity. The FR computation in Section 6.2 is a substantive correctness problem, not just an omitted argument. The abstract advertises the numerical comparison between the ABM and the full-information equilibrium as a main finding, and equation (32) does not compute a Nash equilibrium of the stated FR game because it solves two unrelated games with different rate vectors and splices together one marginal from each. This invalidates the FR curves in Figures 10–12 and the conclusion that ABM outcomes are close to the full-information equilibrium. The fluid model and the discrete-time algorithm for the common-rate model appear sound, so the central theoretical contribution is not overturned; the paper should either solve the coupled FR fixed point explicitly or soften the FR/ABM claims. The verdict therefore remains conditional, with the reason sharpened.","tokens_in":28496,"tokens_out":27311,"duration_ms":287026,"concrete_test":"For the parameters of Figures 10–12, form the coupled best-response operator: given (F_a,F_b), compute BR_a(F_b) with rates ν_a=(8.2,1.8), service distribution z_a, and BR_b(F_a) with rates ν_b=(1.8,8.2), service distribution z_b, using Algorithm 1 with a fine epsilon. Iterate this map to convergence and compare the fixed point with the spliced pair from (32). A simpler sufficient check: evaluate the type-a expected waiting-time vector w_a(t) at the spliced pair using ν_a and z_a; if p̂_e_a assigns positive mass to a slot where w_a(t) is strictly above its minimum, then the spliced pair is not an FR equilibrium.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In Section 6.2 the paper defines the fully-rational (FR) equilibrium but computes it via (p̂_e_a, •) = Alg.2(ν_a,z,epsilon,delta) and (•, p̂_e_b) = Alg.2(ν_b,z,epsilon,delta), taking one component from each run. This is not a fixed point of the two-type best-response map. In the FR Bayesian game a single pair (F_a,F_b) must satisfy: F_a is a best response to F_b when arrival rates are ν_a=(λ α_aa, λ α_ba) and all service times have distribution z_a, while F_b is a best response to F_a when arrival rates are ν_b=(λ α_ab, λ α_bb) and service times have distribution z_b. The two runs in (32) solve two different games with different rate vectors; the discarded components may not match the retained ones, and no consistency condition is checked. Consequently the spliced pair need not satisfy either type's equilibrium conditions, and Figures 10–12 plus the claim that the ABM is close to the FR equilibrium are not supported by the computation presented.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies a bottleneck arrival game in which a Poisson population of customers chooses arrival times to a single-server queue with an acceptance period, and where two customer types hold different beliefs about the service-time distribution. The authors present: (i) a partial characterization of Nash equilibria for exponential service times (Theorem 1 plus a conjecture), (ii) an explicit fluid-limit equilibrium construction (Theorem 2), (iii) a discrete-time formulation with general service times, a best-response characterization (Lemma 5), an iterative algorithm (Algorithm 2) and a Kakutani-based existence claim (Proposition 1), and (iv) an agent-based learning model whose long-run arrival distributions are numerically compared with bounded-rationality and fully-rational equilibria. The paper claims to be the first equilibrium analysis of a bottleneck arrival game with a discrete population and heterogeneous beliefs, and reports that in equilibrium optimistic and pessimistic customers often arrive during disjoint time intervals.","tokens_in":28691,"tokens_out":7993,"duration_ms":80599,"significance":"If the results are correct, the paper makes a useful contribution to strategic queueing by moving beyond homogeneous beliefs: the fluid equilibria of Theorem 2 are given in closed form and provide qualitative insight; Lemma 5 gives a clean fixed-point condition for the discrete game; and the ABM comparison addresses an interesting behavioral question. The authors are also transparent about the main unresolved points, namely Conjecture 1 in the exponential case and the absence of a convergence proof for Algorithm 2. However, the two load-bearing issues described below—the unsupported existence proof in Section 5.2 and the incorrect computation of the fully-rational equilibrium in Section 6.2—currently prevent the paper's central numerical and theoretical claims from being fully established.","major_comments":[{"comment":"The existence proof for the discrete two-type game is not supported. The proposition asserts that the best-response map BR is single-valued and has a closed graph, but neither property is proved. Single-valuedness is not established: uniqueness is only conjectured even in the single-type model of [23], and a fixed-point argument cannot simply assume that the algorithm's output is the unique best response. The claim that 'BR has a closed graph because solutions of (21) are continuous' does not follow, because BR is defined through Algorithm 1, which involves a bisection search with an endogenous support start θ and an endogenous equilibrium payoff w_i; no continuity argument is given for this selection. Since Proposition 1 is the only existence result for the discrete game, the statement that an equilibrium 'exists for any game parameters' is currently unproved. In addition, Algorithm 2 is presented as a method for computing equilibria for general service times, but no convergence guarantee is provided (as Remark 2 acknowledges). This limitation should be stated explicitly in the abstract and in the contributions, rather than presenting the algorithm as a general equilibrium-computation method.","section":"Section 5.2, Proposition 1 and Algorithm 2"},{"comment":"The 'fully rational' equilibrium used for the ABM comparison is not actually computed. In Eq. (32), the authors run Algorithm 2 twice with different rate vectors, taking the first component from the run with νa and the second component from the run with νb. But the FR game is a single coupled fixed-point problem: a pair (F_a,F_b) must satisfy F_a ∈ BR_a(F_b; νa, z_a) and F_b ∈ BR_b(F_a; νb, z_b). The first run computes a Nash equilibrium of an artificial game in which both types have arrival rates νa and service distribution z_a; its second component is not the type-b strategy of the original FR game. Similarly, the second run computes an equilibrium of a different artificial game with rates νb. The spliced pair is not checked for consistency, so it need not be a Nash equilibrium of the two-type FR game. Consequently, the claim that the ABM outcomes are close to the FR equilibrium (Figures 10–12) is unsupported by the computation presented.","section":"Section 6.2, Eq. (32)"}],"minor_comments":[{"comment":"The abstract and introduction say the paper 'characterizes' the Nash equilibrium dynamics for exponential service times, but Theorem 1 is only a partial characterization and the simultaneous-arrival case is left open by Conjecture 1. Please phrase this as a partial characterization or explicitly state that the full characterization depends on an unproved conjecture.","section":"Abstract and Section 3"},{"comment":"In the statement of Theorem 1(i), the expression \"∫_{t_b}^{T} f_b(t) dt = 1−F_b(t)\" appears to contain a typo: the right-hand side should be 1−F_b(0) or 1−F_b(t_b), not 1−F_b(t).","section":"Section 3, Theorem 1(i)"},{"comment":"The formula for the exploration probability θ(x) is malformed as printed: with c_2>0, the expression 1−e^{c_2 x} is negative for x>0, so it cannot define a probability. Please provide the correct formula (likely involving 1+e^{c_2 x}) and specify the parameter values used in the simulations.","section":"Section 6, Eq. (27)"},{"comment":"The notation T is used both for the acceptance period and for its right endpoint (e.g., \"T ⊆ [0,T]\"), which is confusing. Please use a different symbol for one of the two objects.","section":"Section 2"},{"comment":"In the bisection step, when ||p^{(M)}_i|| > 1, the code updates p^{(R)}_{i,θ} := p^{(M)}_{i,θ} but then writes p^{(M)}_{i,θ} := (p^{(L)}_{i,θ} + p^{(R)}_i)/2, where the subscript θ is missing from p^{(R)}_i. Please correct this typo.","section":"Appendix B, Algorithm 1-1"}],"recommendation":"major_revision","confidential_remarks":"The paper has interesting ideas and the fluid-limit analysis is largely sound, but the two major issues are central to the claimed contributions. The Section 6.2 problem is particularly important because the numerical comparison with the ABM is advertised as a main contribution; if the FR equilibrium cannot be computed as described, that comparison needs to be redone with a correct fixed-point computation or removed. The existence proof in Section 5.2 may be repairable, but the authors need either a genuine continuity/single-valuedness argument or a different proof technique. I do not see a circularity problem with the reliance on [23]; that is a legitimate published predecessor. The paper is within scope for the journal, and with the above issues addressed it could become acceptable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The core contribution is real: this is the first bottleneck arrival game with a discrete population and two belief types about service speed, and the fluid equilibrium classification in Theorem 2 is the strongest part. The disjoint-support results and the case-by-case formulas are explicit, checkable, and new relative to the homogeneous-belief literature. The authors are honest about Conjecture 1 being unproven, which I respect. The discrete-time setup with general service times and the fixed-point characterization in Lemma 5 are also useful; Algorithm 2 is a plausible heuristic and the paper correctly says it has no convergence guarantee.\n\nThe soft spots are concentrated in the existence proof and the numerical comparison. Proposition 1's continuity claim for the best-response map is asserted without proof, and it is load-bearing for the Kakutani argument. That is a genuine gap, though not necessarily fatal. Minor but worth noting: Algorithm 1's output depends on a discretization parameter, so the claimed existence is for the approximate algorithm, not the exact game. The bigger problem is Section 6.2. Equation (32) computes the FR equilibrium by splicing one component from two separate Algorithm 2 runs with different rate vectors (νa and νb). That is not a Nash equilibrium of the two-type game because neither type's distribution is a best response to the other type's distribution in the same game. So the claim that the ABM is close to the FR equilibrium is not supported by the computation actually performed. The paper even contains all the ingredients to see this, since νa and νb are different and the posterior service distributions differ. This is a substantive flaw in the numerical section, but it does not undermine the fluid theorem or the single-type best-response machinery.\n\nThe citation pattern is fine: the reliance on [23] is legitimate, and the two-type results go beyond it. No parameter fitting, no invented entities. The paper is earnest and technically careful where it proves things.\n\nBottom line: the fluid analysis alone is worth a serious referee, and the discrete-time model is a useful framework even with the existence gap. The ABM/FR comparison needs either a corrected computation or a much weaker claim. I would send this to peer review with a request for major revision, focusing on Proposition 1 and Section 6.2.","headline":"Genuine first step on heterogeneous-belief arrival games with a solid fluid classification and a clear but incompletely proven discrete-time existence story; the ABM/FR comparison in Section 6.2 is not actually computing a two-type equilibrium.","tokens_in":29201,"tokens_out":811,"would_cite":true,"duration_ms":9787,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60K25","91A10","91A80"],"pacs":[],"model":"deepseek-v4-flash","headline":"In a queue game, beliefs split arrivals into separate time windows.","keywords":["bottleneck queue","strategic arrivals","Nash equilibrium","heterogeneous beliefs","service rate uncertainty","fluid approximation","discrete-time queue","agent-based learning"],"falsifier":"Search the continuous-time exponential model for an equilibrium in which both types' arrival densities are positive on the same interval of times; finding one refutes Conjecture 1 and the uniqueness consequence drawn from it.","tokens_in":28286,"feed_emoji":"🕐","tokens_out":4430,"duration_ms":44809,"temperature":0.7,"pith_summary":"This paper studies when customers choose to arrive at a single-server queue when pessimistic and optimistic customers disagree about the service speed. It tries to establish that, in Nash equilibrium, the two belief types generically arrive during different—often disjoint—time intervals, rather than mixing together. The argument is carried by an explicit fluid-model solution for large systems and by a discrete-time best-response algorithm for general service-time distributions. If the paper is right, bottlenecks with heterogeneous beliefs produce predictable arrival segregation, and mean waiting time grows with the variability of service times.","feed_headline":"Beliefs split queue arrivals into separate time windows","feed_subtitle":"Pessimistic and optimistic customers arrive at disjoint times in equilibrium; the paper gives explicit fluid and algorithmic solutions.","key_machinery":"The machinery is a pair of equilibrium arrival distributions $(F_a,F_b)$ whose supports must satisfy equal-expected-waiting conditions, together with two calculation engines: the fluid-model queue length $q_i(t)=\\sum_j\\lambda_jF_j(t)-\\mu_it$ that yields the explicit Theorem 2 solutions, and the discrete-time mean-workload recursion of Lemma 4, which produces the fixed-point equation (21) that Algorithm 2 solves by iterated best response. The separation of types follows from comparing the two beliefs' expected queue lengths: type-a (pessimistic) customers always face a longer queue than type-b customers for the same arrival profile.","core_discovery":"The central claim is that the bottleneck arrival game with a Poisson population and two belief types—one expecting slow service, one fast—has symmetric Nash equilibria in which the types' arrival distributions are separated in time. For exponential service times the paper characterizes equilibrium structure (Theorem 1) and conjectures no interior overlap (Conjecture 1). In the fluid limit, Theorem 2 gives explicit equilibrium densities for all parameter regimes, including cases where optimistic customers expect zero delay while pessimistic ones face a queue, and cases with multiple equilibria. For general service times, Lemma 5 reduces equilibrium to a system of fixed-point equations, and Algorithm 2 computes equilibria by iterated best response; numerical results show the same separated pattern as the fluid solution. The paper also shows that mean waiting time increases with the coefficient of variation of service time, and that a learning agent-based model approximates the fully informed equilibrium better than the bounded-rationality equilibrium.","pith_inferences":["The signaling mechanism implies that what looks like irrational heterogeneous beliefs can arise from Bayesian updating with noisy signals, so the heterogeneous-belief game can also model a homogeneous Bayesian population; the paper's fully-informed comparison shows this matters for predictions.","If the separation result is robust, a system designer could influence arrival order by strategically releasing information about service speed, steering optimistic customers toward off-peak slots.","The discrete-time algorithm could likely be extended to more than two belief types, but the existence proof via the best-response map would need more care because the continuity assumption may fail in larger type spaces.","The learning-agent comparison suggests that simple reinforcement learning can approximate Nash equilibrium without knowing system parameters, which is a testable hypothesis in controlled queueing experiments."],"forward_implications":["In equilibrium, pessimistic and optimistic customers arrive at different, often disjoint, time intervals; in the fluid model this is proven as Lemma 3, and in the discrete-time numerics it persists.","The fluid model gives explicit arrival densities for every parameter regime, so a planner can predict arrival patterns from the arrival rates, service rates, and acceptance-period length alone.","Mean waiting time increases with the coefficient of variation of service times, so reducing service-time variability lowers average delay in equilibrium.","Customers who only learn from past experience end up with arrival distributions close to the fully rational equilibrium, and can sometimes have lower average waiting times than bounded-rational customers who compute the wrong equilibrium.","Multiple equilibria exist in some fluid parameter regimes, so the arrival game does not always have a unique prediction."],"supporting_citations":[{"why":"Supplies the base ?/M/1 bottleneck arrival game and the homogeneous-customer equilibrium analysis that this paper extends.","marker":"[8]"},{"why":"Establishes the atom-at-zero and continuous-density equilibrium structure for a finite acceptance period that the heterogeneous analysis builds on.","marker":"[11]"},{"why":"Provides the discrete-time equilibrium algorithm and learning model that are extended here to two customer types.","marker":"[23]"},{"why":"Supplies uniqueness machinery and fluid-approximation justification for the bottleneck arrival game.","marker":"[18]"},{"why":"Gives the general-service-time equilibrium form for the homogeneous game that motivates the discrete-time treatment.","marker":"[1]"},{"why":"Provides the join-or-balk model with noisy signals that underlies the paper's information-asymmetry mechanism.","marker":"[6]"}],"fun_headline_variants":["Strategic queue arrivals split by service-speed beliefs","Uncertain service speed splits arrivals by belief","Queue arrivals: your beliefs decide when you show up","Pessimists and optimists queue at different hours","Beliefs create distinct arrival waves in queue games"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument that an equilibrium always exists rests on the best-response map being continuous and single-valued—if at some parameter values the best response to the other type's distribution jumps or fails to be a well-defined distribution, the existence proof and the algorithm's usefulness collapse.","fun_headline_variants_meta":{"raw":{"variants":["Strategic queue arrivals split by service-speed beliefs","Uncertain service speed splits arrivals by belief","Queue arrivals: your beliefs decide when you show up","Pessimists and optimists queue at different hours","Beliefs create distinct arrival waves in queue games"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00098,"raw_usage":{"total_tokens":4186,"prompt_tokens":993,"completion_tokens":3193,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":609,"completion_tokens_details":{"reasoning_tokens":3120}},"tokens_in":609,"tokens_out":3193,"duration_ms":22085,"temperature":1.0,"reasoning_tokens":3120,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:43:47.633532+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Search the continuous-time exponential model for an equilibrium in which both types' arrival densities are positive on the same interval of times; finding one refutes Conjecture 1 and the uniqueness consequence drawn from it.","supporting_citations":[{"cited_title":"Glazer and R","cited_arxiv_id":null,"evidence_quote":"Supplies the base ?/M/1 bottleneck arrival game and the homogeneous-customer equilibrium analysis that this paper extends."},{"cited_title":"Hassin and Y","cited_arxiv_id":null,"evidence_quote":"Establishes the atom-at-zero and continuous-density equilibrium structure for a finite acceptance period that the heterogeneous analysis builds on."},{"cited_title":"Sakuma, H","cited_arxiv_id":null,"evidence_quote":"Provides the discrete-time equilibrium algorithm and learning model that are extended here to two customer types."},{"cited_title":"Juneja and N","cited_arxiv_id":null,"evidence_quote":"Supplies uniqueness machinery and fluid-approximation justification for the bottleneck arrival game."},{"cited_title":"Breinbjerg","cited_arxiv_id":null,"evidence_quote":"Gives the general-service-time equilibrium form for the homogeneous game that motivates the discrete-time treatment."},{"cited_title":"Debo and S","cited_arxiv_id":null,"evidence_quote":"Provides the join-or-balk model with noisy signals that underlies the paper's information-asymmetry mechanism."}],"review_version":1}