{"id":"9d0f6852-f4b5-44fa-bea8-37558ef18fe3","arxiv_id":"2607.23359","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Uniform-index pricing—setting all Weitzman indices equal—is a 4-approximation to optimal revenue for selling information in Pandora's-box search, and is exactly optimal in several special cases.","lead":"This paper shows that a seller pricing access to information in a Pandora's-box search can guarantee at least a quarter of the optimal revenue by charging prices that make the buyer's \"search value\" equal across all options. It also finds several special cases where this simple equal-index pricing is exactly optimal, and studies harder optional-inspection variants.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified to the central 4-approximation theorem; proof is internally sound, with only peripheral issues in Theorem 1 and Theorem 2.","rationale":"The reader's weakest_assumption pointed at tie-breaking and the product-of-CDFs formula. I agree the theorem is convention-dependent, but the paper explicitly adopts seller-favorable tie-breaking and provides an epsilon-perturbation argument, so this is a modeling convention rather than a hidden flaw. The independence assumption is stated up front; correlation is out of scope. Therefore I do not see a load-bearing threat to Theorem 3.\n\nMy independent re-derivation of the theorem's algebra found no error: λ_i is non-increasing, the tail bound goes through the feasible 'removed alternatives' price vector, and the head bound uses λ_k≥1/2 to pass from original indices to g_k. The final uniform-index revenue is evaluated at g_k, so the claimed 4-approximation follows.\n\nThe reader's conditional verdict is still appropriate because the two peripheral issues are legitimate: the MHR theorem is stated without the continuity caveat, and the Theorem 2 numerical search lacks provided code. These do not invalidate the central claim, so I would keep the verdict CONDITIONAL rather than upgrading or rejecting.","tokens_in":32354,"tokens_out":25260,"duration_ms":246136,"concrete_test":"Run an exhaustive enumerator over small discrete instances (n≤3, support size ≤4): for each instance, compute OPT by brute force over all candidate index vectors from Lemma 4, compute max_{g∈S} Rev(I,p(g)), and verify OPT/Rev ≤ 4. Separately compare the two tie-breaking conventions (continue-on-equality vs. stop-on-equality) to confirm the ratio is unaffected. This directly checks the only convention-dependent step in the central proof.","verdict_should_be":"UNCHANGED","load_bearing_attack":"I checked the main proof of Theorem 3 and found no load-bearing flaw. The proof splits OPT at the first index k where the inspection-probability product λ_k = ∏_{j<k} F_j(g_k) falls below 1/2. The tail contribution is bounded by 1/2·OPT via the feasible price vector that removes alternatives 1..k, and the head contribution is bounded by twice the uniform-index revenue at g_k. The required monotonicity of λ_i follows from the sorted indices and independence, both explicit in the model.\n\nThe only convention-dependent step is seller-favorable tie-breaking. Definition 1 formalizes this by maximizing over consistent permutations and continuing on equalities; because all prices are nonnegative, continuing on equality weakly increases revenue, so the formula is coherent. The footnote's ε-perturbation argument shows the convention is implementable up to an arbitrarily small loss. If one insisted on adversarial tie-breaking, the exact statement would need an ε qualifier, but this is not an internal inconsistency.\n\nThe two issues identified by the reader are real but peripheral: Theorem 1(4) assumes continuous MHR (the proof uses E[X|X>0]=E[X]), and Theorem 2 relies on an unverified numerical grid search. Neither affects the central 4-approximation claim.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies revenue maximization by a seller who posts non-adaptive prices for inspecting stochastic alternatives, with a buyer who follows an optimal Pandora's Box search strategy. The central claim is that prices that equalize all Weitzman indices achieve a 4-approximation to the optimal revenue (Theorem 3), and that the best uniform-index price can be computed efficiently for discrete alternatives (Corollary 6). The paper also characterizes exact optimality for several special cases (Theorem 1), shows that uniform prices can be an unbounded factor worse than the optimum (Theorem 8), and develops extensions with seller costs, outside options, matroid constraints, and optional inspection. The main proof of Theorem 3 is clean and internally sound, but I find a significant gap in the proof of Theorem 2, an unsupported continuity assumption in Theorem 1(4), and a missing justification in Claim 18.","tokens_in":32610,"tokens_out":21318,"duration_ms":198133,"significance":"If the results are correct, the paper makes a valuable contribution to algorithmic mechanism design and data pricing: it gives a simple, efficiently computable, parameter-free 4-approximation for revenue in a natural sequential-information-acquisition model, and it provides a sharp separation between index-based and price-based uniform pricing. The proof of Theorem 3 is short and relies only on Weitzman's characterization, which makes the result accessible and likely to be used as a benchmark. The special-case characterizations and the extensions to costs, matroids, and optional inspection broaden the applicability. The main theorem is not affected by the issues I raise, but the paper currently contains an unverifiable numerical proof and an incorrect generality claim in a secondary theorem, so the manuscript needs revision before it meets the standard of a rigorous theory paper.","major_comments":[{"comment":"The proof of the constant-factor gap is not a formal proof as written. The text admits that the lower bound expression L(n,z,α,β,γ,δ) cannot be optimized analytically, and instead states that a Python greedy grid search was run and that L>1.14 for the listed floating-point parameters. No code, no exact arithmetic/interval verification, and no reproducible description of the search are provided. A numerical grid search is not a mathematical certificate, and the reader cannot verify the claimed inequality for the stated parameters. This is a load-bearing gap for the theorem that uniform-index prices are not optimal. Please replace this with an analytic construction or provide a machine-checked interval/rational certificate and the corresponding code.","section":"Appendix A.2, Theorem 2"},{"comment":"The proof says 'as MHR variables are continuous, E[X|X>0] = E[X]'. This is not a consequence of the MHR definition given in the footnote: discrete distributions (e.g., geometric or two-point distributions) can have monotone hazard rate under the standard discrete definition, and the paper uses discrete distributions elsewhere. The proof also needs the assumption that there is no atom at 0 for E[X|X>0] = E[X]. Please either restrict the MHR case to continuous distributions with support not containing an atom at 0, or adapt the argument to discrete MHR distributions. As written, the statement of Theorem 1(4) is broader than what the proof supports.","section":"Section 3, Theorem 1(4) proof"},{"comment":"The proof asserts that ordering alternatives 2..n by decreasing r_i 'ensures that whenever the buyer selects an alternative j≥2, they will select the one of maximum realization.' This is not true for arbitrary distributions with equal indices: once an observed value exceeds μ1, the buyer halts and may never see a later, larger realization. The claim is salvageable for two-point distributions because μ_j ≤ μ1 implies that any value above μ1 is necessarily the high value r_j, and ordering by decreasing r_j then yields the global maximum. But this argument is absent, and the assertion as stated is false in general. Since Claim 18 is used in the proof of Theorem 16, this needs a proper justification in the two-point setting.","section":"Appendix C.3, Claim 18"}],"minor_comments":[{"comment":"In the exchange argument, if F_i(g) = 1 for some alternative, the conditional expectation E[X_i | X_i > g] is undefined. Such alternatives have price 0 and can be handled separately (they never block later alternatives). Please add a sentence to cover this edge case.","section":"Section 2, Lemma 5"},{"comment":"The discussion of existence of a maximizer for unbounded supports is informal ('epsilon-approximate the supremum'). This is acceptable for the main result, but the notation could be tightened by stating all theorems with an explicit ε-optimal price vector for unbounded distributions.","section":"Section 3, footnote 6"},{"comment":"The proof invokes an extension of Theorem 1 to mandatory inspection with an outside option y ≥ 0, saying 'the details are omitted.' Since Theorem 1 as proven in the main text is for y = 0, please include the (short) argument or a reference, so the special-case optimality claim is self-contained.","section":"Appendix C.1, proof of Theorem 15"}],"recommendation":"major_revision","confidential_remarks":"The central 4-approximation theorem (Theorem 3) is in good shape: the proof is sound, self-contained, and does not rely on any fitted constants. The issues are in secondary theorems: Theorem 2's numerical proof is not verifiable as written, Theorem 1(4) contains an unsupported continuity assumption, and Claim 18 needs a missing justification. All appear fixable, but the numerical-search proof of Theorem 2 should be replaced by a certificate or an analytic argument before I would support acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The central result is solid and worth your time. The paper introduces Pandora's Box Pricing—a single seller prices access to box inspections, the buyer follows Weitzman, and the seller maximizes expected revenue. The main theorem, uniform-index prices give a 4-approximation, is clean, and I checked the proof: the splitting at the first k where the inspection probability drops below 1/2 is correct, the tail bound uses the optimal revenue itself, and the head bound goes through with the right monotonicity. Nothing circular or fitted. The support lemma (Lemma 4) and the efficient computation corollary are nice, and the extensions to seller costs and uniform matroids are sensible. The special-case characterizations for i.i.d., two-point, and MHR distributions are also meaningful, and the unbounded gap for uniform prices is a good contrast. Credit where due: the model is new relative to Armstrong and Gamp, and the main proof is genuinely elegant.\n\nThe soft spots are real but minor, and the reader's report has them in the right proportion. First, Theorem 1(4) asserts that MHR variables are continuous, which is false—geometric distributions are MHR and discrete. The proof uses E[X|X>0]=E[X], which depends on continuity at zero. That's a genuine gap in a theorem statement, but it doesn't touch Theorem 3, and I suspect it can be patched with an approximation argument or a slightly different proof. Second, Theorem 2's lower bound (uniform-index not optimal, ratio >1.14) rests on a Python grid search with no code or detailed parameter tables. That's frustrating for reproducibility, but it's a peripheral result. The example itself seems well-formed; it just needs verification. The tie-breaking convention is seller-favorable, which is standard and justified by the epsilon-perturbation footnote; I don't see a load-bearing issue there. The optional inspection section is more exploratory, and the factor n/(n-1) for identical alternatives is clever, but the reduction relies on a fixed outside option and the claims about revenue domination are plausible though not fully airtight.\n\nThis is a paper for people working in algorithmic mechanism design, data pricing, and sequential search. It deserves a serious referee: the main result is strong, the model is natural, and the presentation is clear. I would engage with it and would accept a review invitation. The fixes are small: correct the MHR statement, release the code or convert the grid search into an analytic bound, and tighten the optional-inspection reduction if possible. After that, it should be publishable in a good theory venue.\n\nRecommendation: send it to peer review, and expect minor revision rather than major surgery.","headline":"A natural new problem, a clean and sound 4-approximation theorem, and a couple of small fixable blemishes on the periphery.","tokens_in":33102,"tokens_out":1802,"would_cite":true,"duration_ms":19642,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["91B26","68Q25"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims that for the Pandora's Box Pricing problem, posting prices that equalize the Weitzman indices of all alternatives always yields expected revenue within a factor of 4 of the optimal revenue, and that this simple rule is exa","keywords":["Weitzman index","Pandora's box","posted-price mechanism","revenue maximization","sequential search","approximation algorithm","uniform-index pricing","optional inspection"],"falsifier":"Compute, for a concrete instance with say three independent Bernoulli alternatives, the exact optimal revenue and the best uniform-index revenue using the paper's closed-form expressions; if any instance has OPT > 4 max_g Rev(I,g), Theorem 3 is false. Conversely, finding an instance with ratio above 1.14 would merely tighten the known gap, not refute the approximation claim.","tokens_in":32195,"feed_emoji":"📦","tokens_out":4912,"duration_ms":43110,"temperature":0.7,"pith_summary":"The paper studies a seller who posts a price for inspecting each of several alternatives with privately observed values, while the buyer inspects them sequentially in Weitzman index order and stops optimally. The main result is that a simple, efficiently computable pricing rule—setting prices so that all alternatives share the same Weitzman index—extracts at least a quarter of the optimal revenue in every instance. For several natural special cases (two alternatives, identical distributions, two-point supports, and monotone hazard rate distributions), this uniform-index rule actually achieves the optimum. The paper also shows that equalizing prices themselves, rather than indices, can be a log n factor worse than optimal, and gives positive results for an optional-inspection variant. If true, this means near-optimal information selling does not require complex mechanisms or adaptive pricing.","feed_headline":"A 4-approximation for pricing Pandora's boxes","feed_subtitle":"Posting prices that equalize all Weitzman indices guarantees at least a quarter of optimal revenue from sequential search.","key_machinery":"The Weitzman index g_i of an alternative, defined by p_i = E[(X_i − g_i)^+], is the reservation value that orders the buyer's inspections: the buyer inspects in decreasing g_i and stops once the best observed value exceeds the next index. The paper parameterizes prices by indices and studies uniform-index prices p(g) with all g_i = g; its revenue formula Rev(I,g) = Σ E[(X_i−g)^+] ∏_{j<i} F_j(g) is the central expression analysed in Theorem 3. The support lemma pins the optimal uniform index to the set {0} ∪ ∪_i supp(X_i), turning the search for the best uniform index into a one-dimensional enumeration.","core_discovery":"The central discovery is that the seller's revenue decomposes along the buyer's inspection order: split alternatives at the first index k whose inspection probability drops below 1/2. The revenue from alternatives after k is at most half the optimum, and the revenue from alternatives before k does not shrink by more than a factor of 2 when all their indices are lowered to g_k. Setting the uniform index equal to g_k therefore captures at least a quarter of the optimal revenue, and the proof goes through by monotonicity of the inspection-probability products λ_i = ∏_{j<i} F_j(g_i). The paper further shows the optimal uniform index always lies in the union of the supports of the alternatives, m","pith_inferences":["The gap between the known lower bound (1.14) and the proven upper bound (4) suggests the true worst-case simplicity gap is likely much smaller than 4; a natural next step is to compute the optimal factor by interpolation between uniform-index level sets.","The support lemma implies the optimal uniform index is structurally tied to the alternatives' supports, which may let practitioners fine-tune prices by rounding to observed support points rather than searching over all reals.","For the general optional-inspection model, the paper leaves the arbitrary-distribution case open; its reveal-mechanism technique might extend to distributions with more than two points if the buyer's value ceiling E[max X_i] is used as an upper bound."],"forward_implications":["A seller can compute near-optimal posted prices for arbitrary independent distributions in polynomial time when the distributions are discrete, without solving the buyer's full adaptive problem.","The 4-factor guarantee extends to settings with seller costs, outside options, and uniform-matroid selection constraints, so the uniform-index heuristic is robust beyond the baseline model.","Revenue-optimal pricing in common special cases (identical, two-point, MHR, n=2) coincides with uniform-index pricing, giving exact characterizations of the optimum.","Because uniform prices can be log n worse than optimal, the paper identifies index equalization—not price equalization—as the right simplicity benchmark.","The optional-inspection results (n/(n−1) and 2 approximations) indicate that the hard buyer-side problem can still yield simple near-optimal seller mechanisms in structured cases."],"fun_headline_variants":["Equalize indices, keep 4x revenue","Pandora's box: 4-approx pricing","Simple prices, quarter of optimal","Selling info: 4-approx via index balance","Weitzman equalizer wins 4x revenue"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The revenue formula assumes the buyer breaks ties among equal-index alternatives in the seller's favor; if ties were broken adversarially, or if the alternatives were not independent, the monotonicity of the inspection probabilities—and with it the 4-approximation proof—would fail.","fun_headline_variants_meta":{"raw":{"variants":["Equalize indices, keep 4x revenue","Pandora's box: 4-approx pricing","Simple prices, quarter of optimal","Selling info: 4-approx via index balance","Weitzman equalizer wins 4x revenue"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000173,"raw_usage":{"total_tokens":1158,"prompt_tokens":829,"completion_tokens":329,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":573,"completion_tokens_details":{"reasoning_tokens":256}},"tokens_in":573,"tokens_out":329,"duration_ms":3768,"temperature":1.0,"reasoning_tokens":256,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-31T23:39:07.944640+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute, for a concrete instance with say three independent Bernoulli alternatives, the exact optimal revenue and the best uniform-index revenue using the paper's closed-form expressions; if any instance has OPT > 4 max_g Rev(I,g), Theorem 3 is false. Conversely, finding an instance with ratio above 1.14 would merely tighten the known gap, not refute the approximation claim.","supporting_citations":[],"review_version":1}