{"id":"974d7bd6-b949-49e3-a97d-4461bb9d1c30","arxiv_id":"2505.19453","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"In a duopoly with lotteries, a Stackelberg leader offering one lottery guarantees 1/4 of monopoly revenue for regular distributions, and in the worst case no leader can guarantee more than 1/e.","lead":"Two competing sellers selling an identical good can both profit if one commits first and uses a single lottery: the leader can guarantee a quarter of the monopoly revenue for any regular buyer-value distribution. A companion result shows no leader can guarantee more than 1/e, and in simultaneous-move Nash play both sellers may earn zero even when a monopolist could extract full surplus.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified to Theorem 1.1; the 1/4 bound and its key lemma are internally consistent.","rationale":"The reader's conditional verdict is reasonable: the central Stackelberg approximation theorem is well-supported, but the paper has a secondary gap in Theorem 3.11 and a model limitation in the sequential timing. My read agrees that the timing is a limitation, but I do not see it as a flaw in the argument for the stated theorem; the proof of Lemma 3.2 is detailed and the auxiliary-distribution reduction appears correct. The Theorem 3.11 monotonicity issue is real but does not affect Theorem 1.1. Since no load-bearing concern about the central claim materialized, the verdict should remain unchanged from the reader's CONDITIONAL.","tokens_in":31582,"tokens_out":30786,"duration_ms":274114,"concrete_test":"Implement the auxiliary-distribution reduction for the single-lottery mechanism in Theorem 3.1 across a family of regular and DMR distributions (e.g., uniform, exponential, truncated normal) and numerically optimize Bob's mechanism; verify that the posted price FP_v achieves Bob's maximum revenue and that Alice's resulting revenue is at least Rev(D)/4 in each case.","verdict_should_be":"UNCHANGED","load_bearing_attack":"I read the proof of the central claim in good faith and found no load-bearing flaw. The auxiliary-distribution construction (Eq. 4) correctly reduces Bob's best-response problem to a monopolist problem, and Lemma 3.2 (posted-price best response) is supported by Lemmas 3.7 and 3.8. The integral manipulations and monotonicity arguments in those lemmas check out for regular and DMR distributions, including the handling of the zero-density interval [a+(1-z)s, s] where the virtual-value factor is negative and is bounded via monotonicity of zB. The revenue accounting in Theorem 3.1 (Alice's single lottery with p satisfying Gamma_D(p)=Rev(D)/2, z=1/2, Bob's FP_v) yields Rev(D)/4. The sequential timing is an explicit model assumption acknowledged in Section 5; it limits external applicability but does not invalidate the stated theorem. The unproved monotonicity of Gamma_D^{-1} in Theorem 3.11 is a real gap, but it is non-central to Theorem 1.1.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies a sequential-move duopoly in which two sellers each choose an arbitrary mechanism (equivalently, a pricing function over lotteries) to sell an identical good to a single buyer with a known value distribution. The main result, Theorem 1.1, asserts that in a Stackelberg equilibrium the committed leader (Alice) can guarantee at least Rev(D)/4, the quarter of the optimal monopoly revenue, whenever the distribution is regular or DMR; moreover this is achieved by a single-lottery mechanism. The central structural lemma, Lemma 3.2, states that any best response to such a lottery is a posted price. The paper also gives a 1/e upper bound for point-mass distributions, shows that the factor 4 is tight within the single-lottery class, proves a 1/e approximation when Bob is restricted to posted prices, and demonstrates that the positive result fails for pure Nash equilibria. The technical development is extensive, with the Stackelberg proof built on an auxiliary-distribution reduction (Section 3.2) and a revenue inequality for a single seller (Lemma B.15).","tokens_in":102,"tokens_out":6260,"duration_ms":68927,"significance":"If the central claims hold, this is an interesting and nontrivial contribution: it shows that simple randomized mechanisms can restore a constant-fraction of monopoly revenue in a natural model of seller competition with commitment, and it identifies a clean structural property of best responses to lotteries. The paper is careful about its sequential timing assumption and acknowledges that simultaneous or partial-participation models are not covered. The proof machinery is detailed, with omitted steps supplied in appendices; the auxiliary-distribution construction and the derivation of the 1/4 bound are particularly elegant. The 1/e upper bound and the tightness results give a fairly complete picture of the Stackelberg payoff, modulo the gap in the stated generality of Theorem 3.11 discussed below.","major_comments":[],"minor_comments":[{"comment":"Theorem 3.11 is stated for any distribution D, but the proposed mechanism A in Eq. (7) uses the inverse revenue curve Γ_D^{-1}, and the proof of Lemma B.18 explicitly asserts that Γ_D^{-1}(·) is monotone increasing. This monotonicity is not guaranteed for an arbitrary distribution, and without it the pricing function A may not be proper (convex) or even well defined on the required range. The theorem should either be restricted to distributions for which Γ_D is monotone on the relevant interval (for example, increasing hazard-rate distributions), or the proof must be modified to avoid relying on this assumption. This is not load-bearing for Theorem 1.1, but it is a correctness issue in a stated theorem.","section":"Section 3.6, Theorem 3.11 and Eq. (7)"},{"comment":"In the case where the buyer buys from Alice first, the proof contains the sentence \"we must have B(xB) ≤ xB = 1\", but xB is the buyer's allocation probability from Bob and need not equal 1. This appears to be a typo (likely \"xB ≤ 1\") and should be corrected for clarity.","section":"Section 3.5, proof of Theorem 3.9"},{"comment":"The sentence \"The buyer breaks ties in favor of buying from Bob first, then in favor of larger values of x\" introduces a tie-breaking rule on the allocation quantity that is not stated in Section 2. The relation between this rule and the earlier tie-breaking convention (\"first in favor of maximizing expected allocation from Bob, then from Alice\") should be made explicit.","section":"Section 3.5, line after Eq. (3)"},{"comment":"The model requires the buyer to fully resolve one seller's mechanism before interacting with the other. The paper correctly notes in Section 5 that this is a modeling assumption and that other timing protocols are outside the scope, but it would be helpful to state this limitation near the definition of the timing in Section 2, so that the scope of the main theorems is unambiguous from the outset.","section":"Section 2, Timing"},{"comment":"There are a few typographical errors, e.g., \"equilibirum\" in Section 3.5 and \"whehter\" in the proof of Theorem 3.9. These should be fixed in a final revision.","section":"Throughout"}],"recommendation":"minor_revision","confidential_remarks":"The paper's central Theorem 1.1 appears correct, and the stress-test concern about the monotonicity of Γ_D^{-1} is confined to the side result Theorem 3.11, which is not load-bearing for the main approximation claim. I recommend minor revision: the authors should fix the statement of Theorem 3.11 (or the proof) and address the smaller clarity points. I do not see grounds for rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"One thing to know about this paper: it is a genuine new result in mechanism design for competing sellers, not a repackaging of Bertrand or Myerson. The main theorem gives a 4-approximation to monopoly revenue for a Stackelberg duopoly where the leader can use arbitrary lotteries, and the 1/e upper bound plus the Nash zero-revenue result are real boundary results. The load-bearing piece is the structural lemma: against any single-lottery mechanism for Alice, Bob has a best response that is a posted price. I read through Sections 3.1–3.4 and the appendices carefully; the auxiliary distribution reduction (Lemma 3.5) is elegant and the integral manipulations in Lemma 3.7 and 3.8 hold up for regular and DMR distributions. I could not find a hole in Theorem 1.1 or Lemma 3.2.\n\nThe model assumes the buyer resolves one seller's mechanism completely before engaging the other, with no partial switching. That is an explicit modeling choice, and the authors acknowledge in Section 5 that a simultaneous or interleaved protocol is a natural alternative they don't cover. So the results are conditional on that timing assumption, but it is clearly stated, not smuggled in.\n\nThe real soft spot is Theorem 3.11. It claims a 1/e guarantee for Alice when Bob is restricted to posted prices, for any distribution D. The construction in Eq. (7) uses the inverse revenue curve Γ_D^{-1}, and the proof in Lemma B.18 assumes that inverse is monotone. That monotonicity is not proven and is not true for arbitrary distributions; it holds for regular (or DMR) distributions, where the revenue curve is unimodal. So the statement as written overreaches. This does not affect Theorem 1.1, which is explicitly for regular or DMR, but the claim needs either a regularity assumption or a separate argument for non-regular distributions. The tightness result for single-lottery Alice (Theorem 3.3) is also stated for the point-mass distribution, which is fine but narrower than the title might suggest.\n\nBottom line: this is a solid paper for the algorithmic mechanism design community. The central Stackelberg result is important and, as far as I can tell, correct. I would send it to a serious venue and expect that Theorem 3.11 gets fixed or restated during revision. If you work on mechanism design, it's worth a careful read.","headline":"A new and mostly sound result: constant-factor Stackelberg approximation with lotteries, with one non-central theorem (3.11) that overclaims its domain.","tokens_in":32280,"tokens_out":5302,"would_cite":true,"duration_ms":45434,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["91B26","91A65"],"pacs":[],"model":"deepseek-v4-flash","headline":"A first mover with one lottery secures a quarter of monopoly revenue","keywords":["mechanism design","competing sellers","Stackelberg equilibrium","single-lottery mechanisms","posted-price mechanisms","duopoly revenue","virtual values","approximate revenue maximization"],"falsifier":"Find a regular or DMR distribution and a single-lottery mechanism for Alice such that Bob's revenue-maximizing response beats every posted price; since Lemma 1.2 claims a posted price is always a best response, such an instance would disprove the key structural result. Concretely, on a finely discretized truncated exponential distribution, one could solve the linear program for Bob's optimal mechanism against the half-price lottery and compare its revenue with the best posted price.","tokens_in":31363,"feed_emoji":"🎲","tokens_out":10850,"duration_ms":91904,"temperature":0.7,"pith_summary":"This paper studies two sellers who each choose an arbitrary sales mechanism—possibly including lotteries—to sell identical goods to one buyer who can visit both sellers in sequence. It tries to show that in the Stackelberg version of this game, the first-moving seller can commit to a single-lottery menu and guarantee herself at least one quarter of the revenue a monopolist would earn, for any value distribution that is regular or has decreasing marginal revenue (DMR). The argument rests on a structural lemma: against any single-lottery mechanism, the second seller has a best response that is a posted take-it-or-leave-it price, which turns Alice's design problem into a comparison of fixed prices. The paper also shows the guarantee cannot be pushed above a $1/e$ fraction of monopoly revenue at any Stackelberg equilibrium, and that in simultaneous-move Nash equilibrium both sellers may earn zero even when a monopolist could extract full surplus. If correct, this establishes that randomization together with commitment can overcome Bertrand-style competition, while neither ingredient alone suffices.","feed_headline":"A first mover with one lottery secures a quarter of monopoly revenue","feed_subtitle":"Commitment plus a randomized menu beats Bertrand undercutting; the rival's best reply is just a posted price.","key_machinery":"The load-bearing object is the auxiliary distribution $D_s$ constructed for a fixed threshold $s$, the buyer type that separates those who visit Bob first from those who visit Alice first. When Alice uses a single lottery with price $p$ and allocation probability $z$, the paper shifts the density of types in $[p,s]$ to modified values $a+(1-z)v$, scales down the density above $s$ by $1-z$, and adds a point mass at zero; for any Bob mechanism with threshold $s$, Bob's duopoly revenue equals the revenue a monopolist would earn from $D_s$. This reduction, combined with the standard revenue-as-virtual-welfare identity, shows that Bob's optimal mechanism has allocations concentrated on $\\{0,x\\}$ for types below the threshold and that its revenue is bounded by a convex combination of posted-price revenues, forcing a posted price to be a best response. The proof also introduces 'bottom proper' mechanisms, where every Bob-first type takes the same lottery or nothing, and shows any Bob mechanism can be replaced by a bottom proper one that earns at least as much. The regularity or DMR assumption is exactly what controls the virtual values of the distorted distribution.","core_discovery":"The paper's central claim is Theorem 1.1: for any buyer-value distribution that is regular or has decreasing marginal revenue, there is a Stackelberg equilibrium in which the principal seller, Alice, earns at least $\\mathrm{Rev}(D)/4$, where $\\mathrm{Rev}(D)$ is the maximum expected revenue of a monopolist. The equilibrium is built from a single lottery: Alice charges price $p$ for a probability $1/2$ of the good, where $p$ is the smallest price at which the monopoly revenue curve reaches half its maximum. Against that menu, the paper proves Bob's best response is the revenue-maximizing posted price, and Alice's revenue is at least $\\Gamma(v)/4$ while Bob's is at least $\\Gamma(v)/2$. The companion negative results are Theorem 1.3, which caps Alice's Stackelberg revenue at $\\mathrm{Rev}(D)/e$ even with arbitrary mechanisms, and Theorem 1.4, which shows that with a point-mass buyer type every pure Nash equilibrium gives both sellers zero revenue.","pith_inferences":["Inference: The posted-price best-response lemma is proved only for single-lottery Alice menus; if a similar collapse to posted prices holds for larger menus, the upper bound of $1/e$ might be reachable by a simple first-mover menu, which would close the gap to 4. The paper does not claim this.","Inference: The auxiliary-distribution construction depends on the buyer resolving one mechanism fully before the next; a model that allows interleaving would require a different distortion argument, and the $1/4$ bound may fail there.","Inference: The construction behind the $1/e$ guarantee—a menu that makes Bob indifferent across a range of posted prices—suggests a general template for first movers facing price-undercutting rivals: flatten the rival's best-response revenue over the monopoly price interval.","Inference: Because the positive result needs only regular or DMR priors while the posted-price $1/e$ result needs no regularity, the technical bottleneck is the virtual-value control of the distorted distribution, not competition per se."],"forward_implications":["Alice's quarter-of-monopoly guarantee is achieved by a menu with a single lottery at price $p$ and allocation probability $1/2$, so the optimal first-mover strategy has no menu complexity.","Bob can always be assumed to answer Alice's single-lottery menu with a take-it-or-leave-it price, which reduces a mechanism-design competition to comparison of two prices.","The factor 4 is tight among single-lottery menus for Alice, and the factor $e$ is tight when Bob is restricted to posted prices; the true Stackelberg approximation factor for arbitrary mechanisms lies between $e$ and 4.","Without first-mover commitment, the approximation fails: with a fixed buyer value, every pure Nash equilibrium leaves both sellers with zero revenue.","The $1/4$ guarantee holds for every regular or DMR distribution, and the $1/e$ posted-price result holds for every distribution, even non-regular ones."],"supporting_citations":[{"why":"Supplies the virtual-welfare characterization of revenue that underlies the auxiliary-distribution analysis and the monopoly benchmark.","marker":"[Mye81]"},{"why":"Provides the classic result that a monopolist's optimal single-item mechanism is a take-it-or-leave-it price, defining the benchmark Rev(D).","marker":"[MR78]"},{"why":"Gives the taxation principle that lets each seller's mechanism be represented as a pricing function over allocation probabilities.","marker":"[Ham79]"},{"why":"Complements the taxation-principle argument for strategic equivalence of pricing mechanisms in the two-seller game.","marker":"[Gue81]"},{"why":"Supplies the discrete-distribution analogue of the regularity conditions invoked to extend the results beyond continuous priors.","marker":"[Elk07]"}],"fun_headline_variants":["One lottery wins a quarter of monopoly revenue in Stackelberg duel","First-mover lottery guarantees 1/4 of monopoly revenue","A single lottery secures 25% of monopoly revenue in competition","Stackelberg lottery gets 1/4 of monopoly revenue; rival posts price"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The buyer must finish the first seller's mechanism completely before choosing whether to approach the second; if a buyer could switch mid-mechanism or run the two mechanisms simultaneously, the paper's reduction and its $1/4$ revenue guarantee are not established.","fun_headline_variants_meta":{"raw":{"variants":["One lottery wins a quarter of monopoly revenue in Stackelberg duel","First-mover lottery guarantees 1/4 of monopoly revenue","A single lottery secures 25% of monopoly revenue in competition","Stackelberg lottery gets 1/4 of monopoly revenue; rival posts price"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000594,"raw_usage":{"total_tokens":2801,"prompt_tokens":985,"completion_tokens":1816,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":601,"completion_tokens_details":{"reasoning_tokens":1739}},"tokens_in":601,"tokens_out":1816,"duration_ms":11850,"temperature":1.0,"reasoning_tokens":1739,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T14:14:30.043078+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Find a regular or DMR distribution and a single-lottery mechanism for Alice such that Bob's revenue-maximizing response beats every posted price; since Lemma 1.2 claims a posted price is always a best response, such an instance would disprove the key structural result. Concretely, on a finely discretized truncated exponential distribution, one could solve the linear program for Bob's optimal mechanism against the half-price lottery and compare its revenue with the best posted price.","supporting_citations":[],"review_version":1}