{"id":"41ac8177-1a96-48a3-a21c-3c57ea44ff07","arxiv_id":"2607.29439","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Admission control can increase expected workload when joining decisions depend on observed congestion, but cannot do so over the initial busy period if service requirements and customer types are independent.","lead":"This paper asks whether letting fewer customers into a queue always reduces congestion when customers decide to join based on how congested the queue looks. Using stochastic dominance of workload paths, it shows admission control can backfire, and that backfire disappears if a customer's service need is independent of their patience type.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 2's lower bound rests on a false event claim: a Type-2 arrival in (1.1,1.9) can block the second Type-1, so the proof needs a case analysis (the bound itself appears salvageable).","rationale":"The paper's central positive result, Theorem 3, is supported by a coupling that appears correct under the stated hypotheses (P2)+A(3): the service requirement S_{n_k} is independent of the controlled past because, conditionally on T, S_n is IID with distribution Ψ and independent of Y_n and of earlier join/no-join events. The reader's weakest_assumption about Theorem 3 is therefore not a genuine flaw: the theorem explicitly assumes the independence that concern points to. Counterexample 1 is sound. The one definite defect is in Counterexample 2's proof: the stated event does not guarantee two Type-1 joiners, because an intervening Type-2 can block the second Type-1. This is a real proof error, but it appears repairable by a case analysis, and the lower bound 1.8 likely survives. Since the reader's CONDITIONAL verdict rests on exactly this repair, I do not move the verdict. I also note Section 5.3 defers the periodic-extension proof, but that is an application detail separate from the central claim.","tokens_in":17654,"tokens_out":36686,"duration_ms":382542,"concrete_test":"Partition the event of Theorem 2 Step 2 into E0={no Type-2 arrival in (1.1,1.9)} and E1={at least one Type-2}. On E0, show the first Type-1 in (1,1.1) and the first Type-1 in (1.9,2) both join, yielding \\hat W_2=1.8 exactly. On E1, show the first Type-2 joins and the Type-1 in (1.9,2) balks, yielding \\hat W_2=1.9 (or larger if further joiners occur). If both calculations hold, replace the asserted reason with this case analysis and keep the bound; if E1 can give \\hat W_2<1.8, the lower bound and Theorem 2 fail.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 4.4, Step 2 asserts that on the event that the first arrival in (1,1.1) and the first arrival in (1.9,2) are both Type-1, 'at least two Type-1 customers join' and \\hat W_2^2 ≥ 1.8. This is false as stated. If a Type-2 arrival occurs in (1.1,1.9), it joins (the workload there is 2.9−t ≤1.8<2) and raises the workload at time 1.9 to 2.0, so the Type-1 arrival in (1.9,2) balks. Thus the event does not imply two Type-1 joiners. However, in the worst case one Type-1 plus one Type-2 contributes 1.9 to the workload at time 2, and two Type-1s contribute exactly 1.8, so the lower bound 1.8 appears to survive a partition by whether a Type-2 occurs in (1.1,1.9). This is a proof gap, not a disproof of the theorem. No flaw was found in Theorem 3's coupling or in Counterexample 1.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies whether admission control reduces congestion in a single-server queue where customers' joining decisions depend on the observed workload. It formalizes \"effectiveness\" via stochastic dominance of workload processes (Definition 1) and introduces the weaker notion of initial busy-period effectiveness (Definition 3). The main negative results are two explicit counterexamples (Theorems 1 and 2) in which a policy that suppresses arrivals on [0,1) makes the expected workload at time 2 larger than in the uncontrolled system. The main positive result (Theorem 3) states that if customer type and service requirement are independent (condition (P2)), then every contingent admission control policy is initially busy-period effective; the proof is based on a coupling that transfers to the controlled system the service requirements of customers who joined only the uncontrolled system. An application to stability of state-dependent M/G/1+H queues is sketched in Section 5.3.","tokens_in":17900,"tokens_out":24196,"duration_ms":245238,"significance":"If the results are correct, the paper identifies a genuine behavioral-feedback mechanism through which admission control can backfire, and a clean structural condition (independence of type and service requirement) under which this cannot happen. The counterexamples are simple, explicit, and separate two different mechanisms: a single large service requirement versus the cumulative effect of several small arrivals. Theorem 3 is the main contribution: its coupling proof is self-contained and establishes pathwise dominance, which is substantially stronger than a comparison of means. The stability application shows potential use beyond the immediate setting. The main caveats are that the proof of Theorem 2 contains a false event claim (though the theorem appears salvageable), and the coupling in Theorem 3 is not fully specified at a boundary case, so the central proof needs repair before the results can be accepted as written.","major_comments":[{"comment":"The proof of Theorem 2 claims that on the event that the first arrival in (1,1.1) and the first arrival in (1.9,2) are both Type-1, \"at least two Type-1 customers join\" and that \\hat W_2^2 ≥ 1.8 is in fact an equality. This is false. If a Type-2 arrival occurs in (1.1,1.9), that customer joins (the workload just before that time is 2.9−t_m > 1) and raises the workload at time 1.9 to 2.0, so the Type-1 in (1.9,2) balks. In that path the final workload at time 2 is 1.9, coming from one Type-1 and one Type-2, so the lower bound 1.8 still holds, but the stated event-based reasoning is incorrect. The proof should be repaired by a case analysis that partitions on whether a Type-2 arrival occurs in (1.1,1.9), or by an equivalent argument. This is a load-bearing gap in the second counterexample.","section":"§4.4 Step 2"},{"comment":"The coupling construction for Theorem 3 is not fully specified and appears internally inconsistent. In Step 3, when \\hat V^x_{T_n-}<V^x_{T_n-}, the text assigns \\hat S_n ≡ S_{n_k} unconditionally. But if the number of uncontrolled joiners before T_n equals k−1 (i.e., M(T_n-)=k−1), then the k-th uncontrolled joiner has not yet occurred, so S_{n_k} is not defined at time T_n and may be a future service requirement. Equation (8) in Section 6.2 handles this boundary case by setting \\hat S_n = S_n when M(T_n-)=k−1, which is causally available, but Step 3 does not mention this case. The proof can be repaired by stating and proving the invariant that \\hat V^x<V^x implies M(T_n-)>k−1, so S_{n_k} has already been added to V^x and is available. As written, the central coupling of Theorem 3 is not well-defined, and the pathwise dominance argument requires this missing justification.","section":"§6.1-6.2"}],"minor_comments":[{"comment":"The parameter choice is stated as \"q=λ_h^2\" in the final sentence, but Step 3 defines q(λ_h)=λ_h^{-2}. The displayed formula later should be q=λ_h^{-2}.","section":"§4.4"},{"comment":"The text says \"as illustrated in Figure 2\" when referring to the controlled workload overtaking the uncontrolled one; the correct reference is Figure 3, whose caption is \"Failure of pathwise dominance\".","section":"§3.3"},{"comment":"Typo: \"ˆZ_n is is uniformly distributed\" should read \"ˆZ_n is uniformly distributed\".","section":"§6.2"},{"comment":"The notation W_2^2 (and \\hat W_2^2) in the counterexamples is easy to misread as a square of W_2 rather than the workload at time 2 with initial workload x=2. Consider adding a brief notational reminder, e.g., writing W^2_2(t) or defining the superscript explicitly in the proof.","section":"Notation"},{"comment":"The stability application is only sketched; the proof that condition (5) implies finite expected busy period is deferred to an argument in [7]. Since this is advertised as a contribution, the authors should either provide the details or label the subsection as a sketch.","section":"§5.3"},{"comment":"Reference [12] is listed as an unpublished manuscript. If it is not publicly available, the citation should be marked accordingly or replaced by a published source.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The paper has a valuable core idea and the positive result, once the coupling is made precise, is likely correct. The two major issues — the false event claim in Theorem 2's proof and the underspecified boundary case in Theorem 3's coupling — are repairable locally, but both are load-bearing for the paper's main claims. I would recommend major revision rather than rejection, provided the authors supply the missing case analysis and make the coupling invariant explicit."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Read this one. The paper asks whether admission control can backfire when customers' joining decisions depend on the observed workload, and answers it with two explicit counterexamples plus a clean positive theorem: if service requirements are independent of customer types, any contingent admission control policy is initially busy-period effective (stochastic dominance until first emptiness). The trajectory-based stochastic-dominance criterion for effectiveness is a useful way to frame the question, and the coupling proof of Theorem 3 is the real contribution. It constructs a controlled workload process by matching each admitted customer when the controlled workload is below the uncontrolled one with the service requirement of the next customer who joined the uncontrolled system; under independence of type and service, the marginals check out. I did not find a hole in that argument.\n\nCounterexample 1 is rigorous: a large-service impatient customer can enter after the control period because admission control keeps the workload low, and the controlled workload at time 2 exceeds the uncontrolled one in expectation. Counterexample 2 is the same idea with accumulated small services. But the proof as written has a genuine gap. In Step 2, the event that the first arrival in (1,1.1) and the first arrival in (1.9,2) are both Type 1 does not imply that two Type 1 customers join: a Type 2 arrival in between joins and raises the workload above 1, so the second Type 1 balks. The claimed assertion 'at least two Type-1 customers join' is false. The lower bound W_hat_2 >= 1.8 still survives — replace the second Type 1 with a Type 2 and the contribution is 1 instead of 0.9, giving W_hat_2 = 1.9 — but the proof needs a case split. This is a repair, not a disproof.\n\nThe stability application in Section 5.3 is sketchy: it invokes the same arguments as in Bodas-Jacobovic [7] and drops the periodic-extension details. That is fine for a short paper, but a referee should ask for enough detail to check the regeneration argument. The self-citation is not a problem; it is only used for the auxiliary stability application.\n\nOverall: the math is honest, the main theorem appears correct, and the counterexamples are easy to verify once the gap is patched. The paper deserves a serious referee. I would send it to review with a request to fix Theorem 2's Step 2 and expand the stability sketch slightly.","headline":"A solid counterexample-plus-positive-result paper: the negative examples are real, Theorem 3's coupling works, but Theorem 2's proof needs a small repair.","tokens_in":18404,"tokens_out":5442,"would_cite":true,"duration_ms":53701,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60K25","60K30","68M20","90B22"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper shows admission control can increase congestion in state-dependent queues, and proves that independence between customer types and service requirements restores stochastic dominance of the uncontrolled over the controlled workload","keywords":["admission control","stochastic dominance","workload process","state-dependent behavior","balking","queueing systems","stochastic ordering","feedback effect"],"falsifier":"Simulate the paper's own two-type model but with service times equal to a constant for both types, so that (P2) holds. Use the same suppression policy (no arrivals on [0,1)) and the same initial workload 2. If for any arrival rate and type probability the controlled expected squared workload at time 2 exceeds the uncontrolled one, Theorem 3 is false. The paper's own Theorem 3 predicts the inequality always goes the other way; the search need only scan the two parameters λ_h and q.","tokens_in":17487,"feed_emoji":"🚦","tokens_out":6498,"duration_ms":68520,"temperature":0.7,"pith_summary":"The paper asks whether refusing entry to customers always reduces congestion in a service system. The answer is no when customers observe the workload before deciding whether to join: turning customers away can lower the congestion level just enough to attract later arrivals who would otherwise have balked, and those arrivals can bring either a single very large service demand or a stream of modest ones. The paper constructs two explicit counterexamples in a single-server queue with impatient customers and proves that in both cases the controlled workload exceeds the uncontrolled one at a fixed later time in expectation. The positive result, Theorem 3, identifies when this cannot happen: if each customer's type and service requirement are independent, then any admission-control policy produces a workload path that is stochastically no larger than the uncontrolled path until the system first empties. A sympathetic reader cares because this turns 'admission control works' from an assumption into a condition, and gives a checkable independence condition for when state-aware behavior cannot undo the benefit.","feed_headline":"Blocking arrivals can backfire in queues with state-aware customers","feed_subtitle":"Service needs tied to impatience open the loophole; independence of type and service time closes it.","key_machinery":"The proof of Theorem 3 uses a coupling between the controlled and uncontrolled workload processes. At an arrival where the controlled workload is lower, the controlled customer is assigned the service requirement of the next customer who joined the uncontrolled system — a quantity already contained in the uncontrolled workload but not in the controlled one. Because service requirements are independent of types and history, this swap does not alter the marginal law of the controlled process, and it guarantees that the controlled workload never jumps above the uncontrolled one, yielding pathwise dominance and hence stochastic ordering.","core_discovery":"The central claim is that admission control is not automatically beneficial when joining decisions depend on the workload. The paper constructs two counterexamples in an M_t/G(Ψ)/1 + H(Ψ) queue where suppressing arrivals on an early interval makes the controlled workload exceed the uncontrolled one at a later fixed time — once by admitting a customer whose large service requirement would have been blocked by high congestion, and once by letting several modest customers accumulate. The positive result (Theorem 3) states that when the service requirement of each customer is independent of their type (and of the workload history), every contingent admission-control policy is initially busy-peri","pith_inferences":["The pathwise coupling suggests a stronger, algorithmically useful fact: under independence, even randomized or history-dependent admission policies that admit based on any pre-arrival information cannot make the workload stochastically worse before idleness; one might extend this to controlled release of a queue (e.g., parking arrivals) rather than outright rejection, though the paper does not do ","Real systems with correlated type–service pairs are likely to exhibit the failure; an empirical test would measure the correlation between observed patience and service duration and check whether the intuitive dominance holds when the correlation is near zero.","The busy-period-only result does not address long-run stationary workload; a natural open direction is whether independence also yields stochastic ordering beyond the first busy period when the system is stable — the current coupling argument does not automatically extend past absorption."],"forward_implications":["If the paper is right, designers cannot assume admission control reduces congestion in systems where customers observe the workload and adapt; they must model the feedback loop.","The independence condition (types independent of service requirements) provides a concrete, checkable sufficient condition under which any state-based admission policy is safe up to the first idle period.","The stability application gives a finite-mean busy-period criterion for state-dependent M/G/1+H queues: stable whenever the corresponding state-invariant system with the peak arrival rate is stable, provided service and patience times are independent.","The two counterexamples give explicit templates for when control backfires: when more impatient customers need longer service, or when small arrivals accumulate after a period of suppression."],"fun_headline_variants":["Admission control can backfire when arrivals react to workload","When blocking arrivals worsens congestion: a stochastic ordering analysis","State-dependent customers can make admission control increase congestion","How admission control fails: independence of type and service is key"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The key premise is that a customer's service requirement is independent of their type and of the workload history — if service needs can be predicted from the state a customer observes or from the fact of being admitted, the coupling that sustains the dominance result collapses.","fun_headline_variants_meta":{"raw":{"variants":["Admission control can backfire when arrivals react to workload","When blocking arrivals worsens congestion: a stochastic ordering analysis","State-dependent customers can make admission control increase congestion","How admission control fails: independence of type and service is key"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000218,"raw_usage":{"total_tokens":1233,"prompt_tokens":657,"completion_tokens":576,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":401,"completion_tokens_details":{"reasoning_tokens":510}},"tokens_in":401,"tokens_out":576,"duration_ms":6978,"temperature":1.0,"reasoning_tokens":510,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T07:06:40.216091+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate the paper's own two-type model but with service times equal to a constant for both types, so that (P2) holds. Use the same suppression policy (no arrivals on [0,1)) and the same initial workload 2. If for any arrival rate and type probability the controlled expected squared workload at time 2 exceeds the uncontrolled one, Theorem 3 is false. The paper's own Theorem 3 predicts the inequality always goes the other way; the search need only scan the two parameters λ_h and q.","supporting_citations":[],"review_version":1}