{"id":"8dce5744-8724-4ea0-a3b3-2bbbab1d0221","arxiv_id":"2506.10340","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Under product-quality uncertainty, the optimal number of initial seeds grows logarithmically with network size and should concentrate on a single agent type.","lead":"A startup that knows average connectivity but not the real social network should keep its first seed group small, roughly the log of the population. The paper shows this small group spreads a good product while containing the damage of a bad one.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 1's Θ(log N) bound silently requires λCB(i) − (1−y(i))CG(i) > 0 for every type; when λ is small this fails and the optimal seed count is Θ(N), so the theorem as stated is false.","rationale":"The reader identified the positivity condition A_i > 0 as the weakest assumption; I agree. It is load-bearing because without it Theorem 1's first conclusion is false: the objective is monotone increasing in seeds and the optimum is Θ(N), not Θ(log N). This is not a minor edge case because λ is a free parameter in the model, and any sufficiently small λ violates the condition. The proof's two key formulas—the ER log expression and the IRN q∗ definition—both silently require it. I also considered a second potential issue: the LP's type-selection ratio A_i/[−log(1−y_i)] differs from the threshold ratio A_i/y_i used to define q∗, so the proof may not implement the stated 'lowest marginal cost per probability' criterion exactly. I did not make that the headline because, conditional on A_i > 0, the Θ(log N) scaling survives either way and the discrepancy only shifts the chosen type by a constant; the positivity failure changes the scaling exponent. The proposed test with λ = 0 (or small positive λ) settles the issue: if the exact optimum is N rather than log N, the theorem as stated is false and must be amended. I therefore keep the reader's CONDITIONAL verdict: the intended result is salvageable by adding the positivity assumption and restricting Theorem 1 accordingly.","tokens_in":6830,"tokens_out":16963,"duration_ms":196813,"concrete_test":"Set λ = 0 in the Erdős–Rényi case with any κB < 1 and κG > 1. Because the designer's objective is then AG(S,N), which is nondecreasing in the number of seeds, the exact optimal policy is to seed every agent, giving S∗(N) = N rather than Θ(log N). This directly checks whether the unstated A_i > 0 condition is load-bearing. If the author intends λ to be strictly positive, repeat with a sufficiently small positive λ (e.g., λ = 0.1, κB = 0.5, κG = 2, N = 10^4) and compute the exact expected adoption for S = ⌈log N⌉ and S = N; if the latter yields higher utility, the theorem's Θ(log N) claim fails in that regime as well.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is Theorem 1: for any kernels κG, κB, |S∗(N)| = Θ(log|N|) and it is asymptotically optimal to seed only the type minimizing marginal cost per probability of reaching the giant component. The proof in §3 requires an unstated condition: for every type i, A_i := λCB(i) − (1−y(i))CG(i) > 0. In the Erdős–Rényi derivation, the closed form takes log(D) where D := λ/(1−κB) − (1−y)/(1−(1−y)κG); if D ≤ 0, the inequality holds for all S and the logarithmic formula is invalid. In the IRN relaxation, q∗(N) is defined as max_i {1 − A_i/(y(i)y|N|)}; if A_i < 0, this exceeds 1, and the LP in (1)–(2) has a negative objective coefficient, making the relaxed problem unbounded. The paper never states or proves A_i > 0, and it is not implied by the maintained assumptions λG_1 > 1 and λB_1 < 1, because λ is a free weight. For example, with λ = 0 (the designer values only good-state adoption, which the model does not exclude), A_i = −(1−y(i))CG(i) < 0 for all i, and the objective AG(S,N) is strictly increasing in S, so S∗(N) = N, contradicting Θ(log N). Thus the theorem as stated overclaims; it holds only under an additional parametric restriction that must be made explicit.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies a startup's optimal seeding problem in an inhomogeneous random network when product quality is unknown. The designer chooses a set of initial seeds to maximize expected adoption in the good state minus λ times expected adoption in the bad state. The central theorem claims that, for any kernels, the optimal number of seeds is Θ(log |N|) and that asymptotically it is optimal to seed only the type minimizing the marginal cost per probability of reaching the giant component. The proof develops an Erdős–Rényi special case and a continuous relaxation of the IRN problem, and the paper includes an Instagram back-of-the-envelope illustration.","tokens_in":7193,"tokens_out":10995,"duration_ms":137533,"significance":"If the theorem held as stated, it would offer a clean, parameter-light explanation for why startups launch with small, single-community beta groups rather than mass launches, and it connects usefully to the influence-maximization and seeding literature. The model is transparent, the objective is explicit, and the asymptotic characterization is potentially falsifiable. However, the theorem omits a necessary parametric condition, and the proof actually uses a slightly different optimality criterion from what the theorem states. With the missing assumption added and the criterion stated precisely, the central argument is coherent and the conclusions are defensible.","major_comments":[{"comment":"The proof requires the condition λCB(i) − (1−y(i))CG(i) > 0 for every type i, but this condition is not stated in the theorem or in the model. When the condition fails, the statement is false: for λ = 0 the designer's objective is AG(S,N), which is strictly increasing in S, so S*(N) = N, not Θ(log |N|). In the Erdős–Rényi derivation, D := λ/(1−κB) − (1−y)/(1−(1−y)κG) can be nonpositive, in which case the inequality 'marginal benefit ≥ marginal cost' holds for every S and the displayed ceiling formula for |S*(N)| is invalid. The theorem should be restated under an explicit strict-positivity assumption on these net marginal costs.","section":"§3, Theorem 1"},{"comment":"The LP relaxation is not valid for arbitrary kernels when any coefficient λCB(i) − (1−y(i))CG(i) is nonpositive. If the coefficient is negative, the objective in (1) is unbounded below, so the relaxed problem has no finite solution; if the coefficient is zero, the threshold q*(N) equals 1 and no finite seed set attains the required probability. Thus the claim that the relaxation applies 'for any kernels' is incorrect, and a strict positivity condition is needed for every type.","section":"§3, Eqs. (1)–(2)"},{"comment":"The theorem's prose says the optimal type is the one with the 'lowest marginal cost per probability of reaching the giant component,' but the proof's optimal type j* minimizes [λCB(j) − (1−y(j))CG(j)] / [−log(1−y(j))], not the ratio with y(j) in the denominator. Since −log(1−y) > y for y > 0 and the ordering induced by the two denominators can differ across types, the stated criterion and the proven criterion are not equivalent. The theorem should either use the log-denominator ratio or explain the sense in which 'probability' is meant.","section":"Theorem 1, part 2 vs proof"}],"minor_comments":[{"comment":"The proof of the first part of Theorem 1 is carried out only for a single-type Erdős–Rényi graph; the IRN section presents a relaxation argument that implicitly yields logarithmic scaling but the transition is informal. A more explicit statement of how the ER bound generalizes to the IRN setting would improve rigor.","section":"§3"},{"comment":"In the line defining \\widehat{T}_G, the text has 'i,∈Tj' which should read 'i,j∈T'.","section":"§3"},{"comment":"The illustration uses current Instagram penetration (2/8.1) as an estimate of y, but in the model y is the expected fraction of agents in the giant component conditional on the product being good; these are different quantities. The calibration should be labeled as a rough heuristic rather than a structural estimate.","section":"§3, Instagram illustration"},{"comment":"The statement '|S*(N)| = Θ(log |N|)' is not well-defined when S* is empty, which can occur for small N or large λ. The claim should be qualified as holding for sufficiently large |N| under the maintained assumptions.","section":"Theorem 1"},{"comment":"The phrase 'asymptotically optimal (at rate O(1/|N|))' is imprecise: the proof shows the absolute utility gap is O(1), which is O(1/|N|) only relative to the Θ(|N|) total value. Wording such as 'the per-capita utility loss is O(1/|N|)' would be clearer.","section":"§3"}],"recommendation":"major_revision","confidential_remarks":"The missing positivity condition is a genuine gap in the theorem, but it is repairable by an explicit assumption and by restating the theorem accordingly. The discrepancy between the theorem's prose and the proof's ratio is also fixable. I do not see circularity or parameter fitting; the derivation is from stated primitives, and the Instagram example is not used in the proof. The paper's contribution is potentially solid after these revisions."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: the paper packages a well-known random-graph fact (log-size hitting set for a giant component) into an economic seeding problem with two quality states, and gets a clean answer — seed logarithmically many users of one type. That is a real, useful contribution. The model is simple and the exposition is clear.\n\nThe problem: Theorem 1 as stated ('for any kernels') is false. The proof needs λCB(i) > (1−y(i))CG(i) for every type i — the bad-state cost of a seed that misses the giant component must exceed its good-state small-component benefit. When that fails, e.g. λ=0 (designer values only good-state adoption), the marginal utility of a seed is always positive and the optimum is Θ(N), not Θ(log N). The ER derivation itself takes a log of λ/(1−κB) − (1−y)/(1−(1−y)κG); if that is non-positive the formula is invalid. So the central claim is overstated. This is a load-bearing flaw, but it is mechanically fixable: state the condition, restrict the theorem to the regime where it holds, and explain the economics of the boundary.\n\nWhat else? The IRN part is more heuristic than a full proof — the LP relaxation and the O(1/|N|) gap argument are sketched — but the idea is right and the rounding penalty is small. The Instagram back-of-envelope is cute but it is a post-hoc illustration, not evidence. The comparison to Akbarpour, Malladi, and Saberi is fair, and the discussion of extensions is sensible.\n\nWho is this for? Econ theorists and applied micro people working on seeding, diffusion, and platform launch strategy. It deserves a serious referee: the flaw is serious but patchable, and the intended result is memorable and likely to be cited. I would send it to review, with a request to fix the theorem and discuss the parameter boundary.","headline":"A clean stylized result on why startups seed logarithmically and concentrate on one type, but Theorem 1 only holds under an unstated parameter condition that the paper never proves.","tokens_in":7687,"tokens_out":3423,"would_cite":true,"duration_ms":44021,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["91D30","05C80"],"pacs":[],"model":"deepseek-v4-flash","headline":"The optimal number of initial seeds scales logarithmically with network size and concentrates on a single user type.","keywords":["optimal seeding","diffusion network","inhomogeneous random graphs","giant component","product launch","beta testers","quality uncertainty","logarithmic scaling"],"falsifier":"Take a one-type Erdős-Rényi network with $\\kappa_B < 1 < \\kappa_G$, compute the exact maximizer of $A_G(S) - \\lambda A_B(S)$ for a sequence of populations $N = 10^3, 10^4, \\dots$, and check whether $|S^*(N)| / \\log N$ stays bounded. In particular, choosing parameters that make $\\lambda/(1-\\kappa_B) - (1-y)/(1-(1-y)\\kappa_G)$ nonpositive should produce linear growth, contradicting the theorem if it appears in the region where the stated positivity assumption holds.","tokens_in":6594,"feed_emoji":"🚀","tokens_out":9258,"duration_ms":96448,"temperature":0.7,"pith_summary":"This paper asks how a startup that does not know whether its product is good, and has no map of the diffusion network, should choose its first set of users. The paper argues that the optimal number of initial seeds grows only logarithmically with the size of the population, and that when users come in different connectivity types, all seeds should go to the single type with the lowest marginal cost per probability of hitting the giant component. If true, this explains why many successful products launch inside one small community: a bad product stays localized, while a good product has a high chance of spreading from a few seeds. The result makes the launch decision computable without fine-grained network data.","feed_headline":"Best seed count grows logarithmically with network size","feed_subtitle":"Optimal launch seeds all go to the type with the lowest cost per chance of going viral.","key_machinery":"The machinery is the inhomogeneous random network model, in which each type $i$ connects to type $j$ with probability $\\kappa_X(i,j)/|N|$ in state $X$, and the phase transition is governed by the largest eigenvalue of the kernel matrix. A good product has a supercritical kernel, so a giant component of size $y|N|$ exists; a bad product is subcritical, so all components have size $O(\\log |N|)$. The argument reduces seeding to a linear program by writing the marginal value of a type-$i$ seed as $y(i)[(1-q)y|N|] + (1-y(i))C_G(i) - \\lambda C_B(i)$ and the constraint as a product over seeds; taking logarithms makes the constraint linear, so the optimum is a single type with the lowest ratio $[\\lambda C_B(j) - (1-y(j))C_G(j)] / (-\\log(1-y(j)))$. This ratio is the marginal cost per probability of making the product go viral.","core_discovery":"The central claim is Theorem 1: for any connectivity kernels $\\kappa_G$ and $\\kappa_B$, the optimal seed set satisfies $|S^*(N)| = \\Theta(\\log |N|)$, and it is asymptotically optimal at rate $O(1/|N|)$ to seed only agents of the one type that minimizes the ratio of marginal cost to probability of reaching the giant component. In the single-type case the proof writes $A_G(S)$ as the expected gain from hitting the giant component plus the gain from small components, equates marginal benefit to the constant marginal cost $\\lambda/(1-\\kappa_B)$, and solves for $|S^*(N)|$. In the multi-type case, the integer problem is relaxed to a continuous program whose constraint is a single linear inequality; the relaxed solution seeds only the minimizing type, and rounding the solution loses at most a constant amount of utility, which vanishes at rate $1/|N|$.","pith_inferences":["Editorial inference: when the net value of seeding an off-giant node is positive, the optimal seed count should jump from logarithmic to linear in $|N|$; a simulation that varies $\\lambda$ across that threshold would test the boundary of the paper's main theorem.","Editorial inference: the same single-type ratio should reappear in dynamic settings where product quality improves after adoption, with waves of seeds corresponding to re-optimized versions of the ratio.","Editorial inference: in networks with clustering, hitting the giant component is a less decisive event, so the predicted concentration on one type may weaken; adapting the model to clustered random graphs would be a natural stress test."],"forward_implications":["A doubling of the target population adds only a constant number of extra seeds, so small launches are not a sign of underfunding.","Concentrating all seeds in one community is asymptotically optimal; the startup should not diversify across types to hedge.","The optimal type is identifiable using only type-level quantities $y(i)$, $C_B(i)$, and $C_G(i)$, so the policy is implementable without network-level data.","Adding a constant marginal cost per seed leaves the logarithmic scaling unchanged.","The model's back-of-envelope calculation for a global launch gives about 80 seeds, which matches the observed beta-tester counts of several large platforms."],"supporting_citations":[{"why":"Supplies the inhomogeneous random graph model and its phase transition, which gives the giant-component target and the small-component sizes that drive the marginal costs.","marker":"Bollobás et al. (2007)"},{"why":"Establishes that a constant number of extra seeds can outweigh detailed network knowledge, the comparison point for the paper's claim that seeding strategy needs only type-level information.","marker":"Akbarpour et al. (2023)"},{"why":"The closest treatment of seeding under product-quality uncertainty; the paper contrasts its timing-based contagion mechanism with the network-structure mechanism here.","marker":"McAdams and Song (2025)"},{"why":"Shows that too many seeds can be harmful when adoption depends on neighbor usage, motivating the cost the startup pays when a bad product reaches many users.","marker":"Iyer and Adamic (2019)"},{"why":"Provides tractable seeding rules in networks and frames the general intractability that the paper's type-level logarithmic result avoids.","marker":"Sadler (2025)"}],"fun_headline_variants":["Optimal seed count scales logarithmically","Best seeding: log N seeds, one agent type","Seed logarithmically, target cheapest viral type","Optimal seeding follows log network size","For virality, seed log N and one type"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"For the logarithmic scaling to hold, seeding an extra user who will not reach the giant component must have negative expected value; otherwise the marginal benefit of another seed stays positive after the giant component is already seeded, and the optimal number of seeds becomes linear in network size.","fun_headline_variants_meta":{"raw":{"variants":["Optimal seed count scales logarithmically","Best seeding: log N seeds, one agent type","Seed logarithmically, target cheapest viral type","Optimal seeding follows log network size","For virality, seed log N and one type"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000133,"raw_usage":{"total_tokens":1063,"prompt_tokens":797,"completion_tokens":266,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":413,"completion_tokens_details":{"reasoning_tokens":198}},"tokens_in":413,"tokens_out":266,"duration_ms":3375,"temperature":1.0,"reasoning_tokens":198,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T04:30:43.114812+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a one-type Erdős-Rényi network with $\\kappa_B < 1 < \\kappa_G$, compute the exact maximizer of $A_G(S) - \\lambda A_B(S)$ for a sequence of populations $N = 10^3, 10^4, \\dots$, and check whether $|S^*(N)| / \\log N$ stays bounded. In particular, choosing parameters that make $\\lambda/(1-\\kappa_B) - (1-y)/(1-(1-y)\\kappa_G)$ nonpositive should produce linear growth, contradicting the theorem if it appears in the region where the stated positivity assumption holds.","supporting_citations":[{"cited_title":"Just a Few Seeds More: The Inflated Value of Network Data for Diffusion","cited_arxiv_id":null,"evidence_quote":"Establishes that a constant number of extra seeds can outweigh detailed network knowledge, the comparison point for the paper's claim that seeding strategy needs only type-level information."}],"review_version":1}