{"id":"188d362d-181f-426f-a782-a054bad8bc93","arxiv_id":"1908.07105","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A Bayesian persuasion model shows that an authority can achieve minimal traffic spillover on a route by sending a carefully designed signal to a fraction of travelers above a threshold smaller than one.","lead":"This paper analyzes how a transportation authority can design a noisy signal about road conditions to steer traffic flows in a two-route network where only some drivers receive the signal. It shows mathematically that the authority can achieve the minimum possible overflow on one route as long as the share of informed drivers exceeds a threshold below 100%.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The minimum-spillover claim rests on Proposition 1 and Lemmas 1–3, none proved in this manuscript; the threshold λ is unsupported if those deferred equilibrium and optimality results fail.","rationale":"Reader's weakest assumption matches: Proposition 1 proof omitted and self-cited. I checked the algebra of (11)-(12) against the population-2 indifference condition for the full-information case and the formulas are internally consistent; no numerical contradiction in Example 1. But because the manuscript explicitly says 'we do not present detailed proofs', the global optimality proof (Lemmas 2-3) is not auditable from the text. This justifies the CONDITIONAL verdict: the central insight is plausible and the spot-checks pass, but acceptance should require either a complete proof of Proposition 1 and Lemmas 2-3, or an independent verification (e.g., a companion derivation or machine-checked proof). I do not see grounds to reject or to mark unverified, since the model is well-posed and the formulas are coherent; hence no change to the reader's verdict.","tokens_in":11974,"tokens_out":24624,"duration_ms":225950,"concrete_test":"Independently re-derive Proposition 1 from the equilibrium conditions (8) by solving the complementarity system for q1_r(s) and q2_r for an arbitrary feasible π, without citing [17] or [3]; verify that the only regimes are g(π)≥λ and g(π)<λ, that (11) and (12) are the unique route flows, and that g(π) in (10) equals the maximum possible f*2(a)-f*2(n). In addition, for the parameters of Example 1, numerically grid-search (β,P) satisfying (17) and evaluate (18) for λ∈{0.05,0.1,0.133,0.2,0.25,0.5}; if any feasible (β,P) yields spillover below the Theorem 1 value, the threshold claim is false.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 3's Proposition 1 is the foundation of the paper: it asserts a unique Bayesian Wardrop equilibrium for every feasible π and λ, partitions all information structures into Π1={g(π)≥λ} and Π2={g(π)<λ}, and gives the closed-form flows (11)-(12) used throughout. The proof is not given; the text only says 'The idea of the proof follows Theorem 1 of [17] and Theorem 1-2 in [3]' (both self-citations). Section 4 then states Lemmas 2 and 3 with no proof at all, and Lemma 1 is delegated to [18]. Because (16)-(18) are non-convex and the objective is piecewise linear in equilibrium flows, the claimed global minimum (24) and the threshold λ in (21) depend completely on this unverified characterization. A concrete way the concern could land: if the equilibrium is not unique in some regime, or if there is a third regime (e.g., only one population uses a single route), or if g(π) in (10) does not correctly measure the maximum signal-induced flow change, then the proof of Lemma 2 (which forces π*∈Π1 for λ<λ) and the constancy of spillover for λ≥λ would fail. The τ-restriction (14) and the typo in c^n_1(f1) (state-n cost written with f2, b2) are secondary; they do not affect this core dependency.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper studies a Bayesian persuasion design in a two-route, two-state non-atomic congestion game with partial adoption: a fraction λ of travelers receive a signal about the uncertain network state, while the remaining travelers do not. The authority chooses an information structure π to minimize the average spillover on route 2 at the induced Bayesian Wardrop equilibrium. The main results are Proposition 2, which states that for low accident probability no information is optimal and spillover is zero, and Theorem 1, which gives a piecewise characterization of the optimal information structure in three λ-regimes when the accident probability is high. The central claim is that the minimum possible average spillover is attained for any λ above a threshold λ with λ<1, and Section 5 adds results on equilibrium population costs together with a numerical example.","tokens_in":12240,"tokens_out":14003,"duration_ms":139373,"significance":"If the results are correct, the paper makes a clean and non-obvious contribution: it shows that a designer can achieve the best possible spillover regulation without informing all travelers, and it provides explicit closed-form formulas for the optimal signal and for the critical adoption fraction. The stylized model is clearly formulated, and the paper gives a concrete and internally consistent numerical illustration of the threshold behavior. The main weakness is that the equilibrium characterization and several optimality lemmas are not proved in this manuscript; the central results are therefore plausible and self-consistent but not fully verifiable from the text alone.","major_comments":[{"comment":"Proposition 1 is load-bearing but not proved in this manuscript. The text only says that the idea of the proof follows Theorem 1 of [17] and Theorems 1-2 of [3], and both are prior papers by the same authors. The uniqueness of the Bayesian Wardrop equilibrium, the partition of information structures into Π1 and Π2, and the closed-form flows in (11)-(12) are used directly in the reformulation (16)-(18), in Theorem 1, and in the formula for the threshold λ in (21). A reader cannot verify the central claim without these results. Please make Proposition 1 self-contained, either by giving a full proof or by stating the exact cited result and proving that it applies to the present model.","section":"Section 3, Proposition 1"},{"comment":"Lemmas 2 and 3 are stated with no proof and no proof sketch, yet they are essential for Theorem 1. Lemma 2 asserts that for λ<λ̄ the optimal information structure lies in Π1, and Lemma 3 asserts that f2*(s)≥τ for both signals under the optimal structure. These are not immediate consequences of Proposition 1 because the objective in (16) is non-convex and piecewise linear. The value of the minimum spillover in (24) and the threshold λ in (21) both depend on these lemmas. Section 5's Proposition 3, which underlies the cost comparisons and the claim that λ is also the average-cost-minimizing adoption level, is likewise unproved. Please provide complete proofs or detailed derivations for all of these results, and address explicitly how the non-convexity in (16) is handled.","section":"Section 4, Lemmas 2 and 3; Section 5, Proposition 3"}],"minor_comments":[{"comment":"The expression for c^n_1(f1) is written as α^n_1 f2 + b2, which would make the nominal-state cost on route 1 depend on f2 and equal to the cost structure of route 2. This appears to be a typo; it should presumably be α^n_1 f1 + b1, and it should be corrected.","section":"Section 2.1, cost functions"},{"comment":"The definition of g(π) is typeset ambiguously: it is unclear whether D appears in the denominator of each fraction or only in the first term. Please rewrite (10) with clear parentheses so that the units and the claim g(π)∈[0,1] are unambiguous.","section":"Section 3, Eq. (10)"},{"comment":"In the definition of L(π,f), the formula uses f_r(s), but the loss is the spillover on route 2 only. Please replace f_r(s) with f_2(s), or clarify the notation.","section":"Section 2.2, Eq. (5)"},{"comment":"The introduction says the paper provides the optimal information structure 'for any given spillover threshold', but Theorem 1 is proved only for τ satisfying (14), with the extension for other τ deferred. Please qualify the introductory claim or include the extension.","section":"Section 1 and Section 4"},{"comment":"There are several wording slips, for example 'the fraction λ is induced by players' choice of accessing to the signal versus chosen by the designer versus' in Section 5. A careful proofreading pass is needed.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The manuscript reads like an extended abstract rather than a full journal submission: all central proofs are deferred, and the key equilibrium theorem is imported from the authors' prior work. I recommend requiring a fully self-contained proof appendix before another round of review, and correcting the state-n cost function typo before publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: the partial-adoption spillover result is a real contribution and the three-regime characterization is exactly what people in Bayesian persuasion for routing will want. The threshold λ is non-obvious, and the model is clean: a two-route network, binary state, affine costs, and only a fraction λ of travelers receiving the signal. The finding that the minimum spillover can be achieved with λ < 1 is a concrete, useful insight for city authorities.\n\nWhat's new: unlike Das et al. and Tavafoghi/Teneketzis, this paper lets only a fraction of travelers see the signal and targets spillover on one route. The design of the information structure across the three λ regimes—full revelation in Λ1, a tailored partial signal in Λ2, and a λ-invariant structure in Λ3—is elegant. I spot-checked the formulas in Theorem 1 against the equilibrium expressions in (11)–(12), and they are internally consistent. The numerical example helps illustrate the magnitude of improvement.\n\nSoft spots: the load-bearing Proposition 1 is not proved. It partitions all information structures into Π1/Π2 and asserts unique closed-form equilibrium flows. The text refers to Theorem 1 of [17] and Theorems 1–2 of [3], both by the same authors. For a self-contained paper, that is a gap. The stress-test note is right: if uniqueness fails, or if a third equilibrium regime exists, then Lemma 2 and the threshold λ in (21) collapse. I suspect the proof is straightforward given the two-route affine structure, but it needs to be in the manuscript or an appendix. Also, the cost function in state n is typeset with f2, b2—clearly a typo that makes the model harder to read. The τ-restriction in (14) is stated as a range, and the claim that other cases can be extended is not backed up. Minor: in the definition of Π1/Π2, the set is written as {Π|...} where π is the intended variable.\n\nWho it's for: researchers in transportation information design and Bayesian persuasion with heterogeneous receivers. It deserves a serious referee, because the question is well-posed and the partial-adoption angle is genuinely new. But I would not accept without full proofs of Proposition 1 and Lemmas 2–3, and I would not cite it until the equilibrium lemma is verifiable in the paper itself.\n\nRecommendation: send it to peer review, but with a request for complete proofs and a typo fix.","headline":"A genuine partial-adoption insight for information design in two-route routing games, but the paper leans on unproved equilibrium lemmas and needs to show its work before the main theorem stands alone.","tokens_in":12752,"tokens_out":2781,"would_cite":false,"duration_ms":30904,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["91A26","91A10","90B20"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper shows that a central authority can reach the minimum possible average spillover on a protected route by signaling only a fraction of travelers, provided that fraction exceeds a threshold below 1.","keywords":["Bayesian persuasion","information design","routing game","Bayesian Wardrop equilibrium","traffic spillover","partial information","two-route network","congestion game"],"falsifier":"Compute, for a fixed feasible $\\pi$ and $\\lambda$, all Bayesian Wardrop equilibria of the subgame by solving the best-response inequalities numerically across the parameter range; if any equilibrium flow on route 2 falls outside the two expressions in Proposition 1, or if two different equilibria exist for the same $\\pi$, then Proposition 1 is false and the threshold $\\underline{\\lambda}$ in Theorem 1 does not guarantee the claimed minimum spillover.","tokens_in":11770,"feed_emoji":"🚦","tokens_out":6554,"duration_ms":60713,"temperature":0.7,"pith_summary":"The paper asks whether a central authority can reduce excessive traffic on a chosen route by sending a noisy signal about an uncertain incident, even when only some travelers receive that signal. It claims that, in a two-route network with affine congestion costs and Bayesian Wardrop equilibrium route choices, the optimal information structure attains the minimum possible average spillover on the protected route as soon as the adoption fraction $\\lambda$ exceeds a threshold $\\underline{\\lambda}<1$. When the incident probability is low, the best policy is to reveal nothing, and the spillover is zero. When the incident probability is high, the optimal signal is fully revealing in the nominal state and partially revealing in the accident state, and the spillover floor has a closed-form expression.","feed_headline":"Partial signal to a minority of drivers achieves minimum spillover","feed_subtitle":"In a two-route network with uncertain incidents, the regulator's optimal signal reaches its floor once adoption exceeds a threshold below 1.","key_machinery":"Three tools carry the argument. First, the equilibrium concept is Bayesian Wardrop equilibrium: each traveler chooses the route with minimum expected cost given the signal (for informed travelers) or the prior (for uninformed travelers), and no traveler can lower expected cost by switching. Second, the paper partitions all feasible information structures into two regimes via a function $g(\\pi)$ that compares posterior beliefs against the adoption fraction $\\lambda$; in regime $\\Pi_1$, informed travelers switch routes with the signal and uninformed travelers split, while in regime $\\Pi_2$ both populations split and flows depend on the belief difference induced by the signal. Third, Lemmas 1--3 transform the non-convex design problem into a tractable one: Lemma 1 characterizes which belief-signal pairs $(\\beta,P)$ are feasible, Lemma 2 places the optimum inside $\\Pi_1$ for small $\\lambda$, and Lemma 3 shows the objective is linear because the optimal route-2 flow never falls below the threshold $\\tau$. Theorem 1 then splits $\\lambda$ into three intervals at $\\underline{\\lambda}$ and $\\bar{\\lambda}$ and gives closed-form optimal structures and flows in each.","core_discovery":"The central claim is that in this two-route Bayesian routing game, the authority's optimal information structure achieves the minimum possible average spillover on route 2 as soon as the adoption fraction $\\lambda$ is at least a threshold $\\underline{\\lambda}<1$. For $p\\leq \\bar{p}$, the optimum is to provide no state information, yielding zero spillover. For $p>\\bar{p}$, the paper characterizes the optimal signal piecewise over three regimes of $\\lambda$: in the first regime ($\\lambda<\\underline{\\lambda}$) the signal is fully revealing and spillover decreases with $\\lambda$; in the second and third regimes ($\\lambda\\geq\\underline{\\lambda}$) the spillover is pinned at the floor, and the route-2 flow equals the threshold $\\tau$ in the nominal state and exceeds it only in the accident state. The designer therefore never needs to reach every traveler to reach the best possible outcome.","pith_inferences":[],"forward_implications":["For $p\\leq\\bar{p}$, providing no information is optimal and achieves zero average spillover on route 2.","For $p>\\bar{p}$, the minimum possible average spillover is $\\frac{(D-\\tau)(\\bar{\\alpha}_1(\\theta)+\\alpha_2)-\\alpha_2 D-b_2+b_1}{\\alpha_1^a+\\alpha_2}$, attained for every $\\lambda\\geq\\underline{\\lambda}$.","In the optimal structure, the nominal state is revealed perfectly ($\\pi^*(n|n)=1$), while the accident state is revealed only partially when $\\lambda>\\underline{\\lambda}$, preventing too many travelers from being pushed onto route 2.","The threshold $\\underline{\\lambda}$ decreases as the allowed flow $\\tau$ on route 2 increases, so a city willing to tolerate more flow on the protected route needs fewer informed travelers to reach the spillover floor.","When $\\lambda<\\bar{\\lambda}$, informed travelers enjoy lower average cost than uninformed travelers; if travelers choose whether to adopt the signal, adoption will settle at or above $\\bar{\\lambda}$, equalizing the two populations' costs.","The paper fixes $\\lambda$ as an exogenous parameter; treating $\\lambda$ as a second design variable is a natural extension, and the paper's numerical example suggests that $\\lambda=\\underline{\\lambda}$ minimizes both spillover and average cost while leaving informed travelers better off.","If the same two-regime equilibrium characterization extends to networks with more routes or more states, the qualitative result that a strictly partial adoption rate suffices for the regulator's optimum would likely persist.","For a city negotiating with a navigation provider, $\\lambda$ can be read as the app's market penetration rate; the threshold $\\underline{\\lambda}$ gives a concrete lower bound on reach that still delivers the spillover floor."],"supporting_citations":[{"why":"Supplies the two-regime equilibrium characterization invoked in Proposition 1, giving closed-form route-2 flows for the Bayesian congestion game.","marker":"[17]"},{"why":"Extends the same equilibrium analysis to heterogeneous information environments and is cited as the basis for the uniqueness and closed forms in Proposition 1.","marker":"[3]"},{"why":"Underlies Lemma 1's characterization of which belief-signal pairs $(\\beta,P)$ can be induced by a feasible information structure.","marker":"[18]"}],"fun_headline_variants":["Minimum spillover via signals to a subset of drivers","Optimal routing info reaches floor with adoption below 1","Regulator's best signal needs only a threshold share of drivers","Spillover floor achieved once enough drivers get signals","Bayesian persuasion caps spillover with partial adoption"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is Proposition 1's claim that, for every feasible information structure and adoption fraction, the Bayesian Wardrop equilibrium is unique and takes one of two closed-form flow regimes, and the paper does not prove that proposition here, attributing it to earlier work; if that characterization or its uniqueness fails, the optimality results in Theorem 1 are unsupported.","fun_headline_variants_meta":{"raw":{"variants":["Minimum spillover via signals to a subset of drivers","Optimal routing info reaches floor with adoption below 1","Regulator's best signal needs only a threshold share of drivers","Spillover floor achieved once enough drivers get signals","Bayesian persuasion caps spillover with partial adoption"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000412,"raw_usage":{"total_tokens":2132,"prompt_tokens":947,"completion_tokens":1185,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":563,"completion_tokens_details":{"reasoning_tokens":1107}},"tokens_in":563,"tokens_out":1185,"duration_ms":10063,"temperature":1.0,"reasoning_tokens":1107,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T12:26:33.907532+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute, for a fixed feasible $\\pi$ and $\\lambda$, all Bayesian Wardrop equilibria of the subgame by solving the best-response inequalities numerically across the parameter range; if any equilibrium flow on route 2 falls outside the two expressions in Proposition 1, or if two different equilibria exist for the same $\\pi$, then Proposition 1 is false and the threshold $\\underline{\\lambda}$ in Theorem 1 does not guarantee the claimed minimum spillover.","supporting_citations":[{"cited_title":"Informational aspects in a class of Bayesian congestion games,","cited_arxiv_id":null,"evidence_quote":"Supplies the two-regime equilibrium characterization invoked in Proposition 1, giving closed-form route-2 flows for the Bayesian congestion game."},{"cited_title":"Value of Information in Bayesian Routing Games","cited_arxiv_id":"1808.10590","evidence_quote":"Extends the same equilibrium analysis to heterogeneous information environments and is cited as the basis for the uniqueness and closed forms in Proposition 1."},{"cited_title":"Bayesian persuasion,","cited_arxiv_id":null,"evidence_quote":"Underlies Lemma 1's characterization of which belief-signal pairs $(\\beta,P)$ can be induced by a feasible information structure."}],"review_version":1}