{"id":"fb1a589d-c1ba-47b1-a65a-d6dadc148e98","arxiv_id":"1908.04780","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"The paper proposes incentive mechanisms for distributed estimation with strategic agents, but the proof of the optimal mechanism relies on a condition that does not guarantee the required concavity.","lead":"This paper designs payment rules for crowdsourced estimation tasks, aiming to guarantee a target accuracy at the lowest possible total payment to selfish data collectors. The main optimality theorem has a proof gap, so the general claim is not established; the fallback mechanism is more robust.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The reader's concavity objection does not defeat Theorem 1, since (9) makes q(ξ) increasing and gives a single-crossing maximum. The real load-bearing error is Theorem 2's γ_i sign: E[p_i] = -c_i(ξ_iu), violating individual rationality.","rationale":"The stress-test shows the original reject rationale is not sound: condition (9) is sufficient for global optimality of ξ̃_i via monotonicity of q, so the Nash-equilibrium claim in Theorem 1 is correct even for non-convex costs. The true second derivative need not be negative everywhere; the single-crossing property of q is what matters. However, a real defect exists in Theorem 2's payment constant: the minus sign in (29) makes the expected payment equal to −c_i(ξ_iu), destroying individual rationality. Because the fix is local and the intended mathematical argument is clear, the appropriate disposition is conditional acceptance rather than outright rejection. The paper should correct (29) to read '+' and replace the concavity guarantee in the proof of Theorem 1 with the q-monotonicity/single-crossing argument. No other assumption appears to threaten the central claims.","tokens_in":13141,"tokens_out":18314,"duration_ms":178681,"concrete_test":"Compute E[p_i] under (27)–(29) at the equilibrium profile (30): substitute x̂_ri = x̂_i, y_ri = y_i, ξ_i = ξ_iu, and E[(x̂_i − y_j)^2] = 1/(ξ_x+ξ_iu)+1/ξ_ju. The result is −c_i(ξ_iu), not +c_i(ξ_iu); changing the sign in (29) to '+' recovers the claimed equality and individual rationality. Separately verify Theorem 1 by differentiating U_i with q(ξ) = (ξ_x+ξ)^2 c_i'(ξ): confirm the derivative has the sign of q(ξ̃_i) − q(ξ), proving the global maximum without requiring concavity.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The load-bearing defect is in the fallback mechanism M2, Theorem 2. With γ_i defined by (29), the expected payment at the claimed equilibrium is E[p_i] = γ_i − β_i E[(x̂_i − y_j)^2] = γ_i − β_i(1/(ξ_x+ξ_iu)+1/ξ_ju) = −c_i(ξ_iu). This contradicts the theorem's assertion E[p_i] = c_i(ξ_iu) and violates the individual-rationality constraint in (8): for any c_i > 0 the agent would not participate. The intended argument clearly requires a '+' in (29), but as written Theorem 2 and Proposition 2 are false.\n\nThe reader's stated concern about Theorem 1 is not the load-bearing one. Under (9), q(ξ) = (ξ_x+ξ)^2 c_i'(ξ) is strictly increasing. With β_i = q(ξ̃_i), the derivative of expected utility is U_i'(ξ) = (q(ξ̃_i) − q(ξ))/(ξ_x+ξ)^2. Thus U_i increases up to ξ̃_i and decreases after it, so ξ̃_i is the unique global maximizer on [0, ξ_iu] even when U_i is not concave. Condition (9) is exactly a single-crossing condition, not a concavity condition. Appendix A's claim that the true second derivative (41) must be negative and is implied by (9) is incorrect; (41) does not follow from (9), but the theorem's conclusion survives via the q-monotonicity argument. The proof should be rewritten accordingly.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies a static distributed estimation problem in which N self-interested agents choose an effort level (equivalently, a precision ξ_i), observe a noisy measurement of a Gaussian unknown, and report both a measurement and a local estimate to an estimator. The estimator designs payment mechanisms to implement desired effort and truthful reporting in Nash equilibrium while minimizing expected total payment subject to a global MSE constraint and individual rationality. For cost functions satisfying condition (9), the authors propose mechanism M1 (Theorem 1) that implements the optimal effort profile from problem (13) with expected payment equal to each agent's effort cost. For general cost functions, they propose a fallback mechanism M2 (Theorem 2) that elicits maximum effort from selected agents and reduces the estimator's problem to a binary knapsack problem. Two special cases (quadratic and discrete linear costs) and simulations are provided.","tokens_in":13480,"tokens_out":17691,"duration_ms":169924,"significance":"The paper addresses a timely problem and contains several attractive elements: the formulation in Section II is clean, the reduction of the estimator's optimization to (13) and (11) is useful, and condition (9) is a genuine structural insight, namely a single-crossing condition on (ξ_x+ξ)^2 c_i'(ξ) rather than a concavity condition. If established rigorously, the results would be a valuable contribution to mechanism design for crowdsourced estimation, going beyond the binary-effort models in the prior literature. However, the current manuscript contains a sign error that makes Theorem 2 and Proposition 2 false as stated, a gap in the proof of Theorem 1, and unaddressed issues about the availability of a reference agent and the incentive for truthful measurement reports. These issues are load-bearing but appear fixable in revision.","major_comments":[{"comment":"The sign before c_i(ξ_iu) in (29) is wrong. Substituting the equilibrium strategy (30) into the payment (27) gives E[p_i] = γ_i − β_i(1/(ξ_x+ξ_iu)+1/ξ_ju) = −c_i(ξ_iu), which contradicts the asserted E[p_i] = c_i(ξ_iu) and violates the individual-rationality constraint (8). The intended mechanism requires γ_i = β_i(1/(ξ_x+ξ_iu)+1/ξ_ju) + c_i(ξ_iu). As written, Theorem 2 and Proposition 2 are false.","section":"Section V, Theorem 2, Eq. (29)"},{"comment":"The proof claims that condition (9) guarantees the concavity of E[U_i(ξ_i)] via the negativity of (41). This is incorrect: (41) is −2β_i/(ξ_x+ξ_i)^3 − c_i''(ξ_i) with β_i fixed at c_i'(ξ̃_i)(ξ_x+ξ̃_i)^2, whereas (9) is a pointwise condition on c_i'(ξ_i), so (9) does not imply (41)<0 for all ξ_i. The conclusion of Theorem 1 can nevertheless be recovered because (9) makes q(ξ)=(ξ_x+ξ)^2 c_i'(ξ) strictly increasing, so U_i'(ξ)=(q(ξ̃_i)−q(ξ))/(ξ_x+ξ)^2, which changes sign once at ξ̃_i. The proof should be rewritten along these lines.","section":"Appendix A, proof of Theorem 1, Eq. (41)"},{"comment":"Mechanism M1 requires a reference agent j with positive desired effort, since γ_i contains ξ̃_j^{-1}. The paper does not assume that the optimal solution of (13) has at least two positive components; if exactly one agent is selected, the term ξ̃_j^{-1} is undefined or infinite and the mechanism cannot be implemented. Please add an explicit assumption that at least two selected agents have positive effort, or extend the mechanism to handle the single-selected-agent case.","section":"Section IV-B, Eqs. (23)-(24)"},{"comment":"The payment rules (21) and (27) do not depend on the agent's own reported measurement y_ri, so every value of y_ri gives the same utility. Hence the truthful report y_ri=y_i is only weakly optimal, and the claims that the strategy profile is the 'unique equilibrium in strictly dominant strategies' are false as stated. The authors should either make y_ri incentive-relevant in the payment or explicitly restrict the claims to weak implementation and discuss the resulting equilibrium-selection issue for the fusion rule (4).","section":"Section IV-B, Corollary 1 and Section V, Corollary 2"}],"minor_comments":[{"comment":"The notation φ_i∈(0,1) should be φ_i∈{0,1}, and η_i∈(0,1,...,η_i^m) should be η_i∈{0,1,...,η_i^m}.","section":"Eqs. (10), (11), (20)"},{"comment":"'Without loss of generosity' should read 'without loss of generality'.","section":"Section VI"},{"comment":"In the sentence describing the parameter replacement, the reference to '˜ξ_j' should be ξ_h, since Eq. (29) contains ξ_ju and the honest-agent version should replace that quantity with ξ_h.","section":"Corollary 2"},{"comment":"The assumption should explicitly state the smoothness required for the derivatives in condition (9) and in the proofs, since Assumption 1 as written only asserts positivity and boundedness.","section":"Assumption 1"},{"comment":"The phrase 'if the constraint (9) is satisfied at ξ̃_i' is imprecise because (9) is stated for all ξ_i; the surrounding text should be adjusted to reflect that (9) is a global condition on the cost function.","section":"Appendix A"}],"recommendation":"major_revision","confidential_remarks":"The defects identified are repairable in revision: the sign error in (29) is a localized algebraic mistake, and the proof of Theorem 1 can be fixed with the single-crossing argument described above. I therefore recommend major revision rather than rejection. The authors should also provide a precise statement about the number of selected agents and clarify the status of y_ri as a weakly incentivized report."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's the takeaway. The paper is a genuine contribution to mechanism design for distributed estimation: it handles continuous measurements, general cost functions, and explicitly minimizes expected payments, where prior work was limited to binary tasks or special costs. The structure is sensible—an optimal mechanism when a monotonicity condition holds, and a maximum-effort fallback otherwise—and the reductions to knapsack problems are clean. The writing is clear throughout, and the simulations, though basic, sensibly illustrate the payment-accuracy tradeoff.\n\nNow the soft spots. The reader's stated reason for rejection is that condition (9) doesn't imply concavity of the agents' expected utility. That's true as a statement about the proof, but it's not a fatal flaw. With the chosen beta_i, the derivative of expected utility has a single-crossing form: it's positive below the target effort and negative above it. Condition (9) makes the relevant function q(ξ) strictly increasing, so the target is a global maximizer even without concavity. Appendix A's concavity argument is wrong and should be rewritten, but Theorem 1 survives.\n\nThe real problem is in Theorem 2. With beta_i chosen to make utility increasing in effort, the agent selects maximum effort ξ_iu, and the expected payment at the claimed equilibrium is E[p_i] = γ_i − β_i(1/(ξ_x+ξ_iu) + 1/ξ_ju). Plugging in (29) gives −c_i(ξ_iu), not c_i(ξ_iu). That violates the individual-rationality constraint (8) and makes Theorem 2 false as stated. The fix is trivial—the sign in (29) should be a plus—but it's load-bearing: Proposition 2 and the fallback mechanism are broken until changed.\n\nThere are also minor typos (e.g., \"Without loss of generosity\" in Section VI) and the proof of Corollary 2 is merely sketched, which is fine given it parallels Corollary 1.\n\nWho should read this? Researchers working on crowdsourcing incentives, peer-prediction mechanisms, and the interface of estimation and game theory. The core idea is good, the error is a likely one-character typo, and the paper deserves a serious referee. I'd recommend sending it to review with a request to fix the sign and rewrite the concavity passage in Appendix A. I'd also use it in a reading group—it's the kind of paper where a careful reading finds a subtle bug, and that's a useful exercise.","headline":"A worthwhile mechanism-design paper for crowdsourced estimation whose main theorem survives the reader's objection, but whose fallback mechanism has a sign error that must be fixed.","tokens_in":13938,"tokens_out":3850,"would_cite":true,"duration_ms":36356,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A quadratic payment rule can make self-interested data sources report truthfully at minimum cost.","keywords":["incentive mechanism design","distributed estimation","strategic agents","Nash equilibrium","crowdsourcing","effort elicitation","truthful reporting","knapsack problem"],"falsifier":"Choose a cost function $c_i$ that satisfies (9) but has a region of negative curvature, fix a target $\\tilde{\\xi}_i$, set $\\beta_i$ by (23), and evaluate $U_i(\\xi_i) = \\gamma_i - \\beta_i((\\xi_i + \\xi_x)^{-1} + \\tilde{\\xi}_j^{-1}) - c_i(\\xi_i)$ on $[0, \\xi_{iu}]$. If the maximum over $\\xi_i$ is attained away from $\\tilde{\\xi}_i$, Theorem 1's stated profile is not a Nash equilibrium, and the central claim fails for that cost function.","tokens_in":12965,"feed_emoji":"📡","tokens_out":8655,"duration_ms":75941,"temperature":0.7,"pith_summary":"This paper asks how an estimator can buy data from self-interested agents who choose how carefully to measure and may lie about both their measurement and its accuracy. The authors propose a payment rule that compensates each agent based on how well its reported local estimate predicts another agent's reported raw measurement. They show that when effort-cost functions satisfy a technical condition, the payment parameters can be set so that the desired effort and truthful reporting form a Nash equilibrium, with expected payment to each agent exactly equal to its effort cost. That reduces the estimator's problem to choosing which agents to hire and what accuracy to request, subject to a target global mean-squared error. When the condition fails, a modified version of the same payment rule instead induces maximum effort, preserving feasibility but giving up optimality.","feed_headline":"One payment formula gets truthful reports from strategic sensors","feed_subtitle":"The estimator pays each agent its true effort cost while meeting a target global accuracy, even when effort and reports are private.","key_machinery":"The load-bearing object is the pairwise quadratic scoring rule $p_i = \\gamma_i - \\beta_i(\\hat{x}_{ri} - y_{rj})^2$. Its role is to make agent $i$'s payment depend on how well its reported local estimate predicts another agent's raw measurement, which the agent cannot control. Since $\\min_{\\hat{x}_{ri}} E[(\\hat{x}_{ri} - y_j)^2 \\mid \\hat{x}_i]$ is achieved at $\\hat{x}_{ri} = E[y_j \\mid \\hat{x}_i]$, truthful reporting is the optimal message whenever the effort cost does not enter the message choice. The expected squared error $E[(\\hat{x}_i - y_j)^2] = (\\xi_i + \\xi_x)^{-1} + \\xi_j^{-1}$ makes the payment's sensitivity to effort explicit; $\\beta_i$ is calibrated from the target effort by the first-order condition (23), and $\\gamma_i$ by the individual-rationality condition (24). The same machinery, with $\\beta_i$ chosen large enough, converts the mechanism into a max-effort elicitor. At the estimator level, the mechanism justifies replacing the game-theoretic problem (8) with the deterministic resource-allocation problem (13), or the knapsack problem (11) in the fallback case.","core_discovery":"In the static Gaussian estimation model, the paper claims that the payment $p_i(\\hat{x}_{ri}, y_{rj}) = \\gamma_i - \\beta_i(\\hat{x}_{ri} - y_{rj})^2$ can simultaneously solve both incentive problems that plague crowdsourced estimation: agents choose measurement accuracy (effort) and choose what to report. Because the squared loss makes the conditional expectation $E[y_j \\mid \\hat{x}_i]$ the optimal report, an agent's own accuracy directly improves its expected payment, so effort can be incentivized without observing it. Setting $\\beta_i = c_i'(\\tilde{\\xi}_i)(\\xi_x + \\tilde{\\xi}_i)^2$ makes the estimator-chosen effort $\\tilde{\\xi}_i$ the unique best response when others play their prescribed strategies, and setting $\\gamma_i = \\beta_i((\\xi_x + \\tilde{\\xi}_i)^{-1} + \\tilde{\\xi}_j^{-1}) + c_i(\\tilde{\\xi}_i)$ makes the expected payment exactly the effort cost and keeps participation rational. Thus, under condition (9), the truthful-effort profile is a Nash equilibrium and the estimator's problem collapses to minimizing total cost subject to the precision constraint (13). If (9) fails, the same payment form with a larger $\\beta_i$ pushes agents to their maximum effort, and the estimator solves the binary knapsack problem (11); expected payments equal the maximum-effort costs. The fusion of reports uses the standard inverse-variance weighting, so the global accuracy target is met by construction.","pith_inferences":["The proof's concavity step suggests a practical design check that is stricter than (9): verify $-2\\beta_i / (\\xi_i + \\xi_x)^3 - c_i''(\\xi_i) < 0$ on the whole interval $[0, \\xi_{iu}]$ before deploying M1, or choose $\\beta_i$ to enforce it.","The pairwise scoring structure is a continuous-signal peer-prediction scheme, so the mechanism could be combined with robust Bayesian truth serum ideas to handle agents with correlated private signals rather than conditionally independent Gaussian noise.","In repeated deployments, the estimator could estimate cost functions from observed agent behavior and adapt $\\beta_i$ and $\\gamma_i$ over time, extending the static design to dynamic settings.","The expected-payment guarantee is not a worst-case payment guarantee; a risk-averse estimator or one with a hard budget cap would need an additional payment bound or an ex-ante budget constraint."],"forward_implications":["Under condition (9), the estimator can guarantee a target global MSE at the minimum possible expected total payment, with each selected agent paid exactly its effort cost in equilibrium.","For convex effort costs, condition (9) holds automatically, covering the two worked examples: quadratic continuous costs lead to equal efforts across agents, and discrete per-measurement costs lead to a bounded knapsack problem.","The quadratic cost example gives the closed-form minimum payment $l(\\Sigma_t^{-1} - \\xi_x)^2 / N$ when all $N$ agents are hired.","If one honest agent is available to serve as a reference, the truth-telling profile is a unique equilibrium in strictly dominant strategies, removing the need to worry about equilibrium selection.","When condition (9) fails, the fallback mechanism is still feasible and truthful, but no longer payment-optimal; it solves a binary knapsack selection problem in pseudo-polynomial time."],"supporting_citations":[{"why":"Introduces Bayesian truth serum, the conceptual antecedent for eliciting truthful subjective reports in equilibrium.","marker":"[25]"},{"why":"Establishes the peer-prediction method of paying agents based on agreement with another's report, the template for the quadratic payment rule.","marker":"[26]"},{"why":"The closest prior mechanism-design model that requires ground truth after estimation, which this paper removes.","marker":"[30]"},{"why":"Models effort elicitation in binary tasks with binary effort, the main baseline that this paper generalizes to continuous measurements and continuous effort.","marker":"[31]"},{"why":"Previous incentive-based distributed estimation with strategic sensors, limited to a specific cost function, which this paper extends.","marker":"[33]"},{"why":"Supplies the inverse-variance fusion rule and global MSE decomposition used to state the estimator's accuracy constraint.","marker":"[34]"},{"why":"Provides the dynamic-programming algorithms for the knapsack and bounded-knapsack problems the estimator must solve.","marker":"[35]"},{"why":"Identifies the continuous selection problem as a generalized knapsack with variable coefficients, guiding the reduction to problem (13).","marker":"[36]"}],"fun_headline_variants":["Truthful effort and reports from a single payment rule","One quadratic payment aligns strategic agents' incentives","Pay true cost, get honest data in crowdsourced estimation","Strategic data collectors? This payment scheme demotivates lies","Minimal compensation, guaranteed accuracy via clever payment"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The main proof assumes that condition (9), checked pointwise at every effort level, guarantees that each agent's expected utility is globally concave in effort; in fact the relevant second derivative uses the fixed calibrated constant $\\beta_i$ rather than the moving derivative $c_i'(\\xi_i)$, so non-convex cost functions could admit another, better response away from the desired effort.","fun_headline_variants_meta":{"raw":{"variants":["Truthful effort and reports from a single payment rule","One quadratic payment aligns strategic agents' incentives","Pay true cost, get honest data in crowdsourced estimation","Strategic data collectors? This payment scheme demotivates lies","Minimal compensation, guaranteed accuracy via clever payment"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000482,"raw_usage":{"total_tokens":2432,"prompt_tokens":1043,"completion_tokens":1389,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":659,"completion_tokens_details":{"reasoning_tokens":1313}},"tokens_in":659,"tokens_out":1389,"duration_ms":15369,"temperature":1.0,"reasoning_tokens":1313,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:33:42.776273+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Choose a cost function $c_i$ that satisfies (9) but has a region of negative curvature, fix a target $\\tilde{\\xi}_i$, set $\\beta_i$ by (23), and evaluate $U_i(\\xi_i) = \\gamma_i - \\beta_i((\\xi_i + \\xi_x)^{-1} + \\tilde{\\xi}_j^{-1}) - c_i(\\xi_i)$ on $[0, \\xi_{iu}]$. If the maximum over $\\xi_i$ is attained away from $\\tilde{\\xi}_i$, Theorem 1's stated profile is not a Nash equilibrium, and the central claim fails for that cost function.","supporting_citations":[{"cited_title":"Eliciting informative feed- back: The peer-prediction method,","cited_arxiv_id":null,"evidence_quote":"Establishes the peer-prediction method of paying agents based on agreement with another's report, the template for the quadratic payment rule."},{"cited_title":"Parametric prediction from parametric agents,","cited_arxiv_id":null,"evidence_quote":"The closest prior mechanism-design model that requires ground truth after estimation, which this paper removes."},{"cited_title":"Learning to incentivize: Eliciting effort via output agreement,","cited_arxiv_id":null,"evidence_quote":"Models effort elicitation in binary tasks with binary effort, the main baseline that this paper generalizes to continuous measurements and continuous effort."},{"cited_title":"An incentive-based approach to distributed estimation with strategic sensors,","cited_arxiv_id":null,"evidence_quote":"Previous incentive-based distributed estimation with strategic sensors, limited to a specific cost function, which this paper extends."},{"cited_title":"Distributed estimation,","cited_arxiv_id":null,"evidence_quote":"Supplies the inverse-variance fusion rule and global MSE decomposition used to state the estimator's accuracy constraint."},{"cited_title":"Kellerer, U","cited_arxiv_id":null,"evidence_quote":"Provides the dynamic-programming algorithms for the knapsack and bounded-knapsack problems the estimator must solve."},{"cited_title":"A generalized knapsack problem with variable coefficients,","cited_arxiv_id":null,"evidence_quote":"Identifies the continuous selection problem as a generalized knapsack with variable coefficients, guiding the reduction to problem (13)."}],"review_version":1}