{"id":"385d4759-25b2-4c15-921a-a66a3167eb69","arxiv_id":"2505.18061","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"For i.i.d. valuations in the large-market limit, a fixed price captures at least 0.712 of optimal welfare for one item (tight), and optimal dynamic pricing needs only a constant factor more bidders to match the benchmark.","lead":"This paper shows that in very large markets where buyers' values come from one fixed distribution, a simple fixed price can secure about 71.2% of the optimal seller welfare, up from the known 63.2% worst-case guarantee. It also proves that selling many items erases this large-market advantage, and that optimal dynamic pricing needs only a constant-factor increase in bidders to match the optimal mechanism.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 17's reversed-Weibull branch uses an invalid deficit-ratio inference; Theorem 4(c) is not yet supported.","rationale":"The reader's weakest assumption pointed to Lemma 17's reliance on the cited result [45] that the prophet competitive ratio converges to a nonzero constant. I agree this lemma is the pivotal step for the competition complexity theorem, but the more precise problem is not the existence of a nonzero constant; it is the leap from first-order ratio convergence to the second-order deficit ratio for bounded-support distributions. The paper's own Section 3.1 states that Kennedy-Kertz gives asymptotic approximation equal to one for Gumbel and reversed Weibull, so relying on a nonzero constant is reasonable. However, Theorem 4(c) requires a specific second-order asymptotic (the constant (1-γ) in Lemma 17), and the proof's justification for that asymptotic is a non-sequitur for γ<0. This is a genuine correctness risk in a central result. The 0.712 numerical minimization and case-study ratio are also under-justified, but those are less load-bearing: an exact 0.712 is not necessary for the qualitative improvement over 0.632, whereas the reversed-Weibull competition complexity is a stated theorem. I recommend keeping the conditional verdict, with the concrete test above as a condition for resolving the concern.","tokens_in":37026,"tokens_out":9710,"duration_ms":62092,"concrete_test":"For a reversed-Weibull distribution, e.g., uniform on [0,1] (γ = -1) and F(x) = 1-(1-x)^2 on [0,1] (γ = -1/2), compute G_n via the exact recurrence G_{n+1} = ∫_{G_n}^{ω1} F(u) du (equivalently G_{n+1} = E[max(X,G_n)]) for n up to 10^6. Check whether (ω1-G_n)/(ω1-F^{-1}(1-(1-γ)/(n+1))) converges to 1. If the limit is not 1 for either distribution, Lemma 17 is false for γ<0 and Theorem 4(c) is unsupported; if the limit is 1 for both, the reversed-Weibull branch is empirically corroborated.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The competition complexity theorem (Theorem 4) depends on Lemma 17, which asserts G_n(F) ≈ F^{-1}(1-(1-γ)/(n+1)) for all γ<1. For the reversed-Weibull case γ<0, the proof in Appendix D reduces the problem to showing (ω1-G_{n+1})/(ω1-G_n) ≈ (ω1-E_{n+1})/(ω1-E_n), and justifies this by saying it 'follows from the fact that the competitive ratio of the prophet inequality converges asymptotically to a non-zero constant [45].' This implication is not valid. The cited result [45] concerns G_n/E_n (or G_n/E(M_n)); for bounded-support distributions G_n/E_n → 1 only gives first-order closeness. The second-order deficit ratio (ω1-G_n)/(ω1-E_n) can differ while the first-order ratio converges to 1, and [45] gives no information about this deficit ratio. Theorem 4(c)'s constant (1-γ)(Γ(1-γ))^{1/γ} is derived exactly by combining Lemma 17 with Lemma 13, so if Lemma 17 fails for any γ<0, the reversed-Weibull branch of the constant competition complexity result is unproven. This is load-bearing because Theorem 4's 'constant competition complexity' is a central claim that breaks previously established worst-case impossibility results, and the reversed-Weibull family is an explicit part of the statement.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies fixed-price posted-price mechanisms for i.i.d. valuations under a large-market assumption, meaning the distribution is fixed and the number of bidders n tends to infinity. The authors prove that for distributions in the Fréchet domain of attraction the fixed-price welfare guarantee is at least φ_k(α), which for k=1 is claimed to be at least 0.712, improving the classical 1-1/e bound; for Gumbel and reversed-Weibull domains they prove the guarantee is exactly 1. For the k-unit case they prove a lower bound of 1-1/sqrt(2πk) and an asymptotic tightness result for Pareto(2). They also prove revenue analogues and compute the large-market competition complexity of the optimal dynamic policy as C(F)=(1-γ)(Γ(1-γ))^{1/γ}, yielding constant factors for the three extreme-value families. A case study on eBay Cartier-watch bidding data illustrates the computation of the threshold and the resulting guarantee.","tokens_in":37360,"tokens_out":14992,"duration_ms":159615,"significance":"If the results hold as stated, this is a substantial contribution: it replaces the worst-case fixed-price guarantee 1-1/e with a large-market guarantee of about 0.712, gives a clean asymptotic formula for the k-unit problem, and shows that the competition complexity of optimal dynamic pricing is constant under the extreme-value condition, breaking the previously known unbounded worst-case impossibility when the distribution may depend on n. The main proofs are mostly self-contained and use standard extreme-value and regular-variation tools; the lower bound 1-1/sqrt(2πk) is proved rigorously, and the empirical case study is a useful sanity check. However, several load-bearing claims need additional justification: the reversed-Weibull branch of Lemma 17 relies on an invalid implication, the printed reversed-Weibull formula in Theorem 4(c) has a sign error, and the claimed tightness of the 0.712 constant is asserted without a rigorous proof.","major_comments":[{"comment":"The step where the product [(ω1-G_{n+1})/(ω1-E_{n+1})]·[(ω1-E_n)/(ω1-G_n)] is replaced by 1 is not justified by the cited statement that the prophet-inequality competitive ratio converges to a non-zero constant. For bounded-support distributions, convergence of G_n/E_n to 1 gives only first-order closeness; the ratio (ω1-G_n)/(ω1-E_n) can be asymptotically non-constant while G_n/E_n tends to 1. Since Lemma 17 is used to prove Theorem 4(c), this is load-bearing. Please either provide a direct proof that (ω1-G_n)/(ω1-E_n) is asymptotically constant (for example via Karamata-type arguments as in the surrounding lemmas) or cite a precise result from [45] that establishes the needed deficit-ratio convergence.","section":"Appendix D, proof of Lemma 17, γ<0 branch"},{"comment":"The displayed formula for the reversed-Weibull family has the wrong exponent. Substituting γ=-1/α into the general formula C(F)=(1-γ)(Γ(1-γ))^{1/γ} gives C(F)=(1+1/α)(Γ(1+1/α))^{-α}, not (1+1/α)(Γ(1+1/α))^α. The printed version contradicts Corollary 2(c): for α=2 it gives about 1.18, which is below the lower bound e^{γ*}≈1.781. This must be corrected.","section":"Theorem 4(c)"},{"comment":"The claim that the minimum of φ_1(α) is at least 0.712 and is attained at α*≈1.656 is asserted without proof. Since the statement 'apx_1(F)≥0.712' in Theorem 1(a) depends on this numerical minimization, a rigorous lower bound for the minimum (or a certified interval computation) is needed. In addition, the subsequent claim that this bound is tight and is reached by the Pareto distribution with parameter α* requires proving apx_1(Pareto(α*))=φ_1(α*); the paper proves such an equality only for α=2 in Lemma 8. Either provide the missing proof for general α or state the tightness as a numerical observation.","section":"Section 3.1, after Eq. (4)"}],"minor_comments":[{"comment":"There are recurrent typographical issues with parentheses, e.g. 'E(min{k,B n T ))' should be 'E(min{k,B_n^T})'; these make the displayed derivation harder to read.","section":"Proposition 1 proof"},{"comment":"In the eBay case study, the computation 'competitive ratio at least 3962.5/5400 ≈ 73.3%' needs a definition of the denominator 5400; presumably it is the sample maximum, but this should be stated explicitly so the reader can verify the ratio.","section":"Section 5, Welfare and Revenue Guarantees"},{"comment":"The derivation of φ_1(α) states that the optimum is attained at the smallest non-negative solution of the first-order condition, but the global-maximization argument is not given. This is not a problem for the lower-bound direction if any feasible U is used, but the formula for φ_1 as a maximum needs a short justification or a reference.","section":"Eq. (4) and definition of U*(α)"},{"comment":"The table reports approximate values without indicating which entries are rigorous bounds and which are numerical computations; a footnote distinguishing proved bounds from numerically evaluated constants would improve precision.","section":"Table 1"}],"recommendation":"major_revision","confidential_remarks":"The paper is likely correct in its main ideas, but the reversed-Weibull branch of Lemma 17 and the sign error in Theorem 4(c) are load-bearing and need to be fixed; the 0.712 tightness claim also needs a rigorous proof before the paper can be accepted. I would suggest asking the authors to either supply the missing deficit-ratio argument or weaken the claims accordingly."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a real paper with one strong, carefully proved contribution (the 0.712 fixed-price guarantee and k-unit asymptotics in the Fréchet regime) and one branch of the headline competition-complexity result that is currently unproven. The eBay case study is not load-bearing but is sloppy enough to embarrass the paper.\n\nWhat is new: the large-market fixed-price guarantee improves the classic 1-1/e to 0.712 for distributions in the Fréchet domain, and the k-unit characterization 1-1/sqrt(2πk) is shown tight for Pareto. The competition complexity formula C(F)=(1-γ)(Γ(1-γ))^{1/γ} is elegant and, for Fréchet and Gumbel, the proof via Lemma 17 and Karamata's theorem is convincing. Breaking the earlier unboundedness result under a fixed distribution is a genuinely interesting observation.\n\nWhere it gets soft. Lemma 17 asserts G_n(F) ≈ F^{-1}(1-(1-γ)/(n+1)) for all γ<1. For γ>0 and γ=0 the derivation is fine. For γ<0 the proof reduces to showing the deficit ratio (ω1-G_{n+1})/(ω1-G_n) is asymptotically equal to (ω1-E_{n+1})/(ω1-E_n), and justifies that by saying it follows from [45]'s result that the competitive ratio converges to a nonzero constant. That implication is not valid: [45] gives G_n/E_n → c, which for bounded-support distributions only tells you both deficits vanish; it says nothing about their ratio. You can have G_n = ω1 - b/n and E_n = ω1 - a/n with b≠a while G_n/E_n → 1. So Theorem 4(c) is, as written, unsupported. A serious referee should ask the authors to prove the deficit-ratio convergence directly, or restrict Theorem 4(c) appropriately.\n\nAlso the claim that the 0.712 bound is attained at α*≈1.656 is only backed by a numerical minimization; the paper does not provide a proof that this is the global minimum. That matters because the abstract advertises 0.712 as the tight guarantee. And the eBay section's 'competitive ratio of at least 3962.5/5400' compares the threshold to the realized maximum of the sample, not to the expected maximum used in the theorem. The number 73.3% is therefore not a valid instance of the theoretical guarantee.\n\nThe citation pattern is clean, the proofs are mostly self-contained, and the Fréchet analysis is technically solid. I'd send it to a good referee with the request to focus on Lemma 17 and the numerical tightness. If the reversed-Weibull branch can be fixed, this is a strong paper.","headline":"Genuine large-market results in the Fréchet case; the reversed-Weibull branch of the competition complexity theorem rests on an invalid inference and needs repair.","tokens_in":37870,"tokens_out":5080,"would_cite":true,"duration_ms":40517,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60G70","60G40","91B26"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that in a large market with a fixed valuation distribution, a single posted price attains at least 71.2% of the optimal mechanism's welfare, and that the competition complexity of dynamic pricing is a universal constant.","keywords":["posted pricing","fixed price policies","large markets","competition complexity","extreme value theory","prophet inequalities","welfare guarantees","multi-unit auctions"],"falsifier":"A concrete calculation: for the Pareto distribution with shape $\\alpha\\approx 1.656$, the fixed-price welfare ratio for a single item should approach about $0.712$ as $n\\to\\infty$; a limit below $0.71$ would refute the main guarantee.","tokens_in":36862,"feed_emoji":"🏷️","tokens_out":9716,"duration_ms":74004,"temperature":0.7,"pith_summary":"This paper proves that when the market is large—the buyer distribution stays fixed while the number of buyers grows—simple posted prices perform far better than classic worst-case guarantees. For a single item, the guaranteed fraction of the optimal mechanism's welfare rises from the universal $1-1/e\\approx 0.632$ to a tight $0.712$ for distributions in the Fr\\'echet family, and to $1$ for Gumbel and reversed-Weibull types. For $k$ identical units the guarantee is at least $1-1/\\sqrt{2\\pi k}$, so the large-market advantage shrinks as $k$ grows. The paper also proves that the competition complexity—the multiplicative extra bidders a posted-price policy needs to match the optimal mechanism—is a universal constant, at most $e\\approx 2.718$, overturning worst-case unboundedness.","feed_headline":"One fixed price secures 71.2% of optimal welfare","feed_subtitle":"As markets grow, posted pricing approaches optimal welfare; the needed extra bidders stays below e.","key_machinery":"The load-bearing device is the extreme value condition, the analogue of the central limit theorem for the maximum of an i.i.d. sample: it says that $(M_n-b_n)/a_n$ converges to one of three named limit laws. The proofs rescale the price threshold by the quantile $a_n=F^{-1}(1-1/n)$ and pass to the limit; normalized order statistics converge to Poisson-type expressions, and regular-variation asymptotics give the tail integrals. This reduces the fixed-price approximation factor to the explicit optimization $\\varphi_k(\\alpha)$ and reduces the dynamic policy sequence $G_n(F)$ to the quantile approximation $G_n(F)\\approx F^{-1}(1-(1-\\gamma)/(n+1))$, from which the exact competition-complexity constant follows.","core_discovery":"The central discovery is that the extreme value type of the valuation distribution completely governs both problems in the large-market limit. If $F$ has Fr\\'echet type with shape $\\alpha>1$, then for one item the fixed-price welfare guarantee is $\\varphi_1(\\alpha)\\ge 0.712$, with the bound approached by a Pareto distribution of shape $\\alpha^*\\approx 1.656$; for $k$ items, the guarantee is $\\varphi_k(\\alpha)\\ge 1-1/\\sqrt{2\\pi k}$, and this is asymptotically tight as $k\\to\\infty$. If $F$ has Gumbel or reversed-Weibull type, a fixed price asymptotically attains the full welfare of the optimal mechanism. For the optimal dynamic policy, the large-market competition complexity is $C(F)=(1-\\gamma)(\\Gamma(1-\\gamma))^{1/\\gamma}$ with $\\gamma=1/\\alpha,0,-1/\\alpha$ respectively, which lies between $1$ and $e$; this breaks the previously established worst-case impossibility of unbounded competition complexity.","pith_inferences":["If the same extreme-value analysis applies when the market size is random rather than fixed, the constant bidder-inflation factor would give a practical rule: attract a fixed multiplicative extra number of bidders instead of designing item-specific prices.","The asymptotic tightness for $k$ units suggests that large-market gains are concentrated in thin markets with small $k$; platforms selling many units per listing should not expect the same benefit from sheer scale.","A natural testable extension is whether the 0.712 threshold also holds for correlated valuations or for markets where the number of buyers depends on realized prices.","The contrast between welfare results (best for Gumbel) and competition-complexity results (best for Fr\\'echet) hints that the distributions for which simple pricing is easiest are the ones for which matching the optimum by dynamic pricing is hardest."],"forward_implications":["For a fixed Fr\\'echet-type distribution, a single anonymous price recovers at least 71.2% of optimal welfare in a large market, beating the 63.2% guarantee that is tight when the distribution may vary with the market size.","In Gumbel and reversed-Weibull markets, fixed prices asymptotically achieve 100% of optimal welfare (and, under tail-regularity conditions, 100% of optimal revenue), so price discrimination buys nothing in the limit.","For $k$ identical units, the large-market fixed-price guarantee is at least $1-1/\\sqrt{2\\pi k}$ and is asymptotically tight, so the advantage of a large market disappears as the number of units grows.","The large-market competition complexity is a constant between 1 and $e$; in particular, multiplying the number of bidders by $e^{\\gamma_\\star}\\approx 1.781$ suffices for Gumbel distributions, in sharp contrast to the unbounded worst case.","The adaptivity gap—the loss from using fixed prices instead of the optimal dynamic policy—is at most about 1.105 in large markets."],"supporting_citations":[{"why":"Establishes the $1-1/e$ fixed-price welfare guarantee and its tightness that the large-market $0.712$ improves.","marker":"[23]"},{"why":"Supplies the asymptotic value of the optimal dynamic policy, whose positive-constant competitive ratio is used in Lemma 17 to pin down $G_n(F)$.","marker":"[45]"},{"why":"Provides the extreme-value theorem, tail representations, and moment convergence results used in the proofs of Theorems 1 and 3.","marker":"[53]"},{"why":"Gives the $1-1/\\sqrt{2\\pi k}$ approximation ratio for the $k$-unit problem that the large-market lower bound matches asymptotically.","marker":"[21]"},{"why":"Establishes the worst-case unbounded competition complexity that the large-market constant characterization overturns.","marker":"[9, 10]"},{"why":"Supplies the regular-variation and order-statistic limit tools used to derive the quantile asymptotics behind Theorem 4.","marker":"[35]"}],"fun_headline_variants":["Fixed price welfare bound jumps to 0.712 in large markets","Large markets make posted pricing competition complexity constant","Fixed price guarantee rises to 71.2% as market scales","Large markets turn worst-case posted pricing impossibility to constant"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof leans on a cited theorem that the optimal dynamic policy keeps a strictly positive fraction of the maximum valuation in every distribution with an extreme value limit; if that fraction were zero or failed to converge, the constant competition-complexity factors would collapse.","fun_headline_variants_meta":{"raw":{"variants":["Fixed price welfare bound jumps to 0.712 in large markets","Large markets make posted pricing competition complexity constant","Fixed price guarantee rises to 71.2% as market scales","Large markets turn worst-case posted pricing impossibility to constant"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000516,"raw_usage":{"total_tokens":2520,"prompt_tokens":977,"completion_tokens":1543,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":593,"completion_tokens_details":{"reasoning_tokens":1476}},"tokens_in":593,"tokens_out":1543,"duration_ms":12315,"temperature":1.0,"reasoning_tokens":1476,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T14:36:20.928293+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A concrete calculation: for the Pareto distribution with shape $\\alpha\\approx 1.656$, the fixed-price welfare ratio for a single item should approach about $0.712$ as $n\\to\\infty$; a limit below $0.71$ would refute the main guarantee.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes the $1-1/e$ fixed-price welfare guarantee and its tightness that the large-market $0.712$ improves."},{"cited_title":"P., and Kertz, R","cited_arxiv_id":null,"evidence_quote":"Supplies the asymptotic value of the optimal dynamic policy, whose positive-constant competitive ratio is used in Lemma 17 to pin down $G_n(F)$."},{"cited_title":"Revenue Gaps for Static and Dynamic Posted Pricing of Homogeneous Goods","cited_arxiv_id":"1607.07105","evidence_quote":"Gives the $1-1/\\sqrt{2\\pi k}$ approximation ratio for the $k$-unit problem that the large-market lower bound matches asymptotically."}],"review_version":1}