Persuasion under the Threat of Verification

Georgy Lukyanov; Samuel Safaryan

REVIEW 2 major objections 3 minor 1 cited by

Persuasion under the Threat of Verification

T0 review · 2 major / 3 minor · reviewed 2026-08-05 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Cheaper fact-checking makes the optimal public signal less informative, the paper argues.

desk verdict The paper's central comparative static is vacuous because its own Eq. (3.5) makes full revelation optimal, and the key lemma is false. read the letter →

arxiv 2508.19682 v3 pith:2XINJI2E submitted 2025-08-27 econ.TH

Georgy Lukyanov , Samuel Safaryan This is my paper · ORCID

classification econ.TH

keywords BayesianpersuasioninformationdesignverifiableevidencecostlyverificationpublicsignalsBlackwellinformativenessfalsificationrepression

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper studies a sender who commits to a public experiment before a mass audience decides whether to pay a private cost to verify the state. It tries to establish a reverse comparative static: when verification becomes cheaper in the population, the sender's optimal public signal becomes strictly less informative, because extreme claims invite more scrutiny. The mechanism is a constrained concavification problem: folding receivers' endogenous verification into the sender's payoff makes the indirect value more concave in the public belief, so the optimal experiment coarsens. With an ex-post truthfulness constraint on verifiable evidence, cheaper verification also shifts the sender toward ex-post manipulation—upward falsification first, then fixed-cost repression—once thresholds are crossed. If true, the result connects transparency, message precision, and repression in one framework.

What carries the argument

The carrying object is the sender's indirect payoff at a public posterior, v(µ;F) = −(b² + (1−F(µ(1−µ)))²µ(1−µ)), where F(µ(1−µ)) is the mass of receivers who verify because the private benefit of verification equals the posterior variance. The argument works by showing that a first-order stochastic improvement in F makes v(·;F) pointwise more concave, then applying Bayesian-persuasion concavification: the optimal experiment is a Bayes-plausible distribution over posteriors maximizing E[v(µ;F)], and more concavity pulls the supporting chord inward. The ex-post truthfulness constraint enters through the indifference condition (1−F(µ_s(1−µ_s)))µ_s = 2b, which determines the only interior poste

What would settle it

Test the paper's key concavity lemma by choosing a distribution F′ that first-order stochastically dominates F and evaluating v(µ;F′)−v(1/2;F′) against v(µ;F)−v(1/2;F) at an interior belief such as µ=0.1. The paper's conclusion that optimal experiments become Blackwell-less-informative requires this inequality at every µ; a single reversal—for example with F uniform and F′ piecewise linear with F′(0.09)=0.15, F′(0.25)=0.25—would falsify the lemma. One could also directly compute the optimal experiment under each F and check whether the spread of the posterior distribution shrinks.

Watch

Extended reading notes

Core claim

The paper's central claim is that in a public-signal persuasion game where receivers can verify at heterogeneous costs and the sender must be truthful about verifiable claims, a fall in verification costs makes the sender's indirect payoff more concave in the public posterior and thereby makes every optimal public experiment less informative in the Blackwell order. The authors derive this through a concavification argument: the sender maximizes the expected value of v(µ;F) over Bayes-plausible posterior distributions, subject to an ex-post implementability constraint that pins down the silence posterior by (1−F(µ_s(1−µ_s)))µ_s = 2b. Under a protocol with hard evidence plus a minimal soft lab

Load-bearing premise

The load-bearing premise is that a first-order improvement in the verification-cost distribution makes the sender's payoff more concave at every interior belief—and that an interior experiment can beat full revelation, since equation (3.5) alone would make the endpoints optimal.

Editorial extensions

If this is right

Cheaper fact-checking makes optimal public communication coarser: measured by the Blackwell order, the public experiment becomes less informative as the verification-cost distribution improves.
At the optimal message, a larger share of receivers chooses to verify, but the sender's response is to make bold claims rare, so realized verification can fall even as the underlying cost falls.
Under upward-only or capacity-limited falsification, cheaper verification increases the sender's ex-ante use of falsification in the unfavorable state while decreasing it in the favorable state.
There is a fixed-cost threshold for repression: cheaper verification expands the set of posteriors and states in which violence is worth buying.
A policy that lowers verification costs should therefore expect noisier messages, more manipulation, and—at thresholds—discrete repression, rather than a simple increase in transparency.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The same concavification logic implies that the coarsening response should be strongest for receivers whose beliefs sit near 1/2, where the verification benefit µ(1−µ) peaks; this is a testable cross-sectional prediction the paper does not spell out.
When falsification is unconstrained, the persuasion margin is predicted to be neutral—only manipulation scales—so observed message coarsening alongside a verification-cost shock can be used to detect whether falsification capacity binds.
The threshold structure of repression implies a temporal ordering—noisier messages and rising manipulation before a discrete onset of violence—which event studies of censorship breaks or fact-checking rollouts could trace.
A dynamic extension of the model would predict that sustained declines in verification costs produce gradual coarsening and falsification punctuated by repression spikes once accumulated gaps clear the fixed cost.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The setup is genuinely novel: a mass audience that self-selects into costly verification, combined with an ex-post truthfulness constraint, is a nice idea that isn't in the cited literature. The verification cutoff and the aggregate-action derivation are clean, and the falsification/repression extension could be interesting in a different model. But the central result collapses under inspection of the paper's own value function.

Eq. (3.5) gives v(µ;F) = −b² − (1−λ)²µ(1−µ). Since (1−λ)²µ(1−µ) ≥ 0, v is at most −b², with equality only at µ=0 and µ=1. Therefore the sender's expected payoff is maximized by the fully revealing experiment, regardless of F. Any interior posterior reduces the payoff. So the optimum is full revelation for every verification-cost distribution, and Theorem 4.2's claimed strict coarsening under cheaper verification is vacuous—the optimal experiment doesn't change at all. The paper's description of v as 'single-peaked with a global maximum at µ∈{0,1/2,1}' is also wrong; v is actually higher at the endpoints than at 1/2.

The second leg is Lemma 4.1, which asserts that a FOSD improvement in verification costs makes v more concave in a pointwise sense. The proof only shows that (1−λ)²µ(1−µ) pointwise decreases, which does not imply the stated contraction around 1/2. A concrete counterexample exists: take F uniform and a piecewise-linear F' with F'(0.09)=0.15 and F'(0.25)=0.25. This F' is a FOSD improvement (cheaper costs), but at µ=0.1 the inequality in Lemma 4.1 is reversed. So the lemma is false, and the proof of Theorem 4.2 relies on it.

There is also a mismatch with the abstract: it promises an explicit counterexample showing that FOSD improvements alone don't sign the response, but no such counterexample appears in the main text.

The flaws are load-bearing, not cosmetic. The model's objective makes full revelation the unique optimum, so the entire comparative static and the phase-reversal narrative are unsupported. The framework might be salvageable with a different sender objective (e.g., one that rewards belief dispersion), but as written the paper should be rejected. A referee would quickly identify these issues, so the paper does deserve a serious referee to document them, but it should not be published without major revision.

Referee Report

2 major / 3 minor

Summary. The paper studies a sender who commits to a public experiment before a continuum of receivers who can privately verify the state at heterogeneous costs. The sender's loss is quadratic in the distance between the aggregate action and θ+b, and verifiable reports must be truthful (EPIC). The main claim is a reverse comparative static: if the verification-cost distribution improves in FOSD, the sender's indirect value v(µ;F) becomes more concave and every optimal public experiment becomes strictly less Blackwell informative. After the benchmark, the paper imposes EPIC (Protocol A: hard evidence + silence; Protocol B: minimal soft layer) and extends the framework to falsification and repression. The appendix provides proofs and a constructive implementation.

Significance. The paper is clearly written and the EPIC-constrained concavification formulation is a natural way to model hard-evidence constraints in mass-audience persuasion. The closed-form uniform-cost special case is a useful pedagogical benchmark. However, the central comparative static is invalidated by the paper's own Eq. (3.5): the unconstrained optimum is full revelation for every F, and Lemma 4.1, which the proof invokes, is false. As a result, the claimed reverse comparative static — the paper's main contribution — does not follow from the model as specified.

major comments (2)

[Section 4.1, Eq. (3.5), Theorem 4.2] Eq. (3.5) shows v(µ;F)=−b²−(1−λ(µ;F))²µ(1−µ)≤−b², with equality only at µ∈{0,1} for any F with λ(µ;F)<1 on (0,1). Therefore, for any prior π, the fully revealing public experiment achieves the upper bound in problem (4.1), and any distribution with interior support is strictly worse. The unconstrained optimal experiment is the same full-revelation distribution for every F, so Theorem 4.2's claimed strict coarsening is impossible. Under EPIC with ε>0, the optimal policy is δ0=δ1=1 for every F; informativeness does not respond to F.
[Lemma 4.1 / Lemma A.2] The asserted mean-preserving contraction inequality is false. Let F(x)=x on [0,1], and let F' be a cdf with F'(0.09)=0.15 and F'(0.25)=0.25 (a FOSD improvement in the sense of cheaper verification). At µ=0.1, using Eq. (3.5) with b=0, p(µ;F)=(1−F(µ(1−µ)))²µ(1−µ), we have p(0.1;F)=0.074529, p(0.5;F)=0.140625, so v(0.1;F)−v(0.5;F)=0.066096. For F', p(0.1;F')=0.065025, p(0.5;F')=0.140625, so v(0.1;F')−v(0.5;F')=0.0756>0.066096, reversing the inequality. The lemma also mischaracterizes v: at b=0, v(0)=v(1)=0>v(1/2)=−(1−F(1/4))²/4, so 1/2 is a local minimum, not a global maximum. Since Theorem 4.2's proof and Theorem 4.4 rely on this lemma, the coarsening conclusion lacks support.

minor comments (3)

[Section 3.4] The statement that a FOSD decrease in F makes v(µ;F) 'more concave in µ' is asserted without proof and is false as stated; see the counterexample in the major comments.
[Section 4.2, Protocol A] The paragraph after Proposition 4.3 says the informativeness effect is a priori ambiguous, but in the actual model with the quadratic objective the maximal-disclosure policy dominates any non-degenerate protocol, so the ambiguity is not present.
[Section A.8 and Section 6.5] The continuous-state extension is stated without proof ('details are omitted for brevity'). This unsupported claim goes beyond the binary-state analysis and should be either proved or clearly labeled as conjectural.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the derivation is self-contained and the main theorem is a mathematical consequence of stated primitives, not an unpacking of its own inputs.

full rationale

The paper's central claim is the comparative static that a FOSD improvement in the verification-cost distribution makes the sender's indirect payoff v(mu;F) more concave and, consequently, every optimal public experiment less Blackwell-informative. The chain of derivation is explicit and does not rely on any fitted parameter, self-citation, or imported uniqueness theorem. v(mu;F) is derived from primitives in Eq. (3.5): v(mu;F) = -(b^2 + (1-lambda(mu;F))^2 mu(1-mu)). Lemma 4.1 then proves a pointwise contraction property from the fact that F' FOSD F implies lambda(mu;F') >= lambda(mu;F), which is a direct consequence of the cutoff rule (3.2)-(3.3). Theorem 4.2 uses the standard Kamenica-Gentzkow concavification characterization, an external and independently established benchmark, to translate this contraction into a mean-preserving contraction of the optimal binary experiment. There is no step where the conclusion is assumed in the definition of a premise: the 'more concave' property is not defined in terms of Blackwell informativeness, and the Blackwell ordering is applied only after the concavification argument. The paper contains no self-citations at all, so patterns 3-5 do not apply. No parameters are estimated from data and then 'predicted'; all quantities are closed-form functions of primitives, so patterns 1-2 do not apply. The reviewer's concern that Eq. (3.5) implies full revelation is the unique unconstrained optimum, and the alleged counterexample to Lemma 4.1, are substantive correctness objections about whether the assumptions hold and whether the theorem's scope is vacuous; they are not instances of circularity. A false or misstated lemma, if present, is an error in the derivation, not a reduction of the result to its own inputs. The derivation is therefore self-contained, and no circular step can be exhibited with the required precision.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The central claim rests on (i) quadratic losses for both sender and receivers, (ii) the EPIC implementability restriction pinned by Eq. (A.2), and (iii) the asserted curvature property of v in Lemma 4.1. Items (i)-(ii) are domain assumptions; item (iii) is ad hoc and false in general, and is essentially the result itself. No free parameters or invented entities are introduced.

assumptions (5)

domain assumption Quadratic receiver loss gives the private value of verification equal to posterior variance µ(1−µ), producing the cutoff rule verify iff c_i ≤ µ(1−µ) (Eq. 3.2).
Standard in Bayesian persuasion with quadratic loss; it makes the verifying mass λ(µ;F) = F(µ(1−µ)).
domain assumption Sender loss is (A−(θ+b))^2, yielding the indirect value v(µ;F) = −(b^2 + (1−λ)^2 µ(1−µ)) (Eq. 3.5).
This is the primitive that makes the value function single-peaked with endpoints at −b^2.
domain assumption EPIC implementability reduces to δ1=1 and (1−λ(µs;F))µs = 2b (Lemma A.5, Eq. A.2), so the θ=1 branch is fully disclosed and the θ=0 interior posterior is pinned by an indifference condition.
Modeling choice; the paper uses this to generate a constrained design problem.
ad hoc to paper The mean-preserving contraction property of v in Lemma 4.1 holds for all FOSD-improved F.
This is the load-bearing curvature assumption. It is asserted and proved with an invalid argument; a concrete counterexample (uniform F and piecewise-linear F' with F'(0.25)=0.25, F'(0.09)=0.15) violates the inequality at µ=0.1.
ad hoc to paper The optimal experiment is interior and non-degenerate, so the comparative static of the optimal experiment is meaningful.
The paper implicitly assumes the optimum is not full revelation, but Eq. (3.5) implies v(µ)≤−b^2, so the unique optimum is full revelation; the claimed coarsening is vacuous.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Persuasion under the Threat of Verification." pith.science (2026). https://pith.science/paper/2XINJI2E

@misc{pith2026250819682,
  author       = {Pith},
  title        = {Pith review of: Persuasion under the Threat of Verification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2XINJI2E}},
  note         = {Machine review of arXiv:2508.19682}
}

read the original abstract

Public communication is often followed by private fact-finding. We study a sender who commits to a costly public experiment before heterogeneous receivers decide whether to pay to verify the state, and ask whether cheaper private verification disciplines the sender or instead lets her shift the informational burden onto receivers. The answer turns on the sender's own cost of public information. When that cost is low, cheaper verification makes the optimal experiment weakly more informative; once the discipline margin is active, the favourable signal becomes more decisive and realized verification falls. When public information is costly, the sender instead tolerates more private fact-finding, weakly coarsens the experiment, and may eventually pool. A benchmark with uniformly distributed verification costs and quadratic persuasion costs delivers this phase reversal in closed form. For general symmetric convex persuasion costs, primitive curvature and supporting-line conditions recover each side of the reversal; an explicit counterexample shows that a first-order stochastic reduction in verification costs alone does not sign the response. Public informativeness and private verification, in short, need not move together.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

Audit Silence and the Capacity Trap
econ.TH 2025-09 unverdicted novelty 6.0 of 10

A finite run of audit silence can force every sequential equilibrium into a region of maximal firm violation and maximal inspector effort, so the distribution of enforcement capacity matters.

Reference graph

Works this paper leans on

20 extracted references · 20 canonical work pages · cited by 1 Pith paper

[1]

and Babichenko, Y

Arieli, I. and Babichenko, Y. (2019). Private Bayesian persuasion.Journal of Economic Theory, 182:185–217

work page 2019
[2]

Ben-Porath, E., Dekel, E., and Lipman, B. L. (2019). Mechanisms with evidence: Commitment and robustness.Econometrica, 87(2):529–566

work page 2019
[3]

and Morris, S

Bergemann, D. and Morris, S. (2016). Information design, bayesian persuasion, and Bayes correlated equilibrium.American Economic Review: Papers & Proceedings, 106(5):586–591

work page 2016
[4]

and Tirole, J

Caillaud, B. and Tirole, J. (2007). Consensus building: How to persuade a group.American Economic Review, 97(5):1877–1900

work page 2007
[5]

and Skreta, V

Doval, L. and Skreta, V. (2022). Mechanism design with limited commitment.Econometrica, 90(4):1463–1500

work page 2022
[6]

Gehlbach, S., Luo, Z., Shirikov, A., and Vorobyev, D. (2021). A model of censorship, propaganda, and repression. Working paper

work page 2021
[7]

Grossman, S. J. (1981). The informational role of warranties and private disclosure about product quality.The Journal of Law and Economics, 24(3):461–483

work page 1981
[8]

and Treisman, D

Guriev, S. and Treisman, D. (2019). Informational autocrats.Journal of Economic Perspec- tives, 33(4):100–127

work page 2019

Show all 20 references

[9]

and Treisman, D

Guriev, S. and Treisman, D. (2020). A theory of informational autocracy.Journal of Public Economics, 186:104–114

2020
[10]

Kamenica, E. (2019). Bayesian persuasion and information design.Annual Review of Economics, 11:249–272

2019
[11]

and Gentzkow, M

Kamenica, E. and Gentzkow, M. (2011). Bayesian persuasion.American Economic Review, 101(6):2590–2615. 29

2011
[12]

Kolotilin, A., Mylovanov, T., Zapechelnyuk, A., and Li, M. (2017). Persuasion of a privately informed receiver.Econometrica, 85(6):1949–1964. Matysková, L. and Montes, A. (2023). Bayesian persuasion with costly information acquisition. Journal of Economic Theory, 211:105678

2017
[13]

Milgrom, P. R. (1981). Good news and bad news: Representation theorems and applications. The Bell Journal of Economics, 12(2):380–391

1981
[14]

and Shin, H

Morris, S. and Shin, H. S. (2002). Social value of public information.American Economic Review, 92(5):1521–1534

2002
[15]

Seidmann, D. J. and Winter, E. (1997). Strategic information transmission with verifiable messages.Econometrica, 65(1):163–170

1997
[16]

Shin, H. S. (1994). News management and the value of firms.RAND Journal of Economics, 25(1):58–71

1994
[17]

Shin, H. S. (2003). Disclosures and asset returns.Econometrica, 71(1):105–133

2003
[18]

and Pérez-Richet, E

Skreta, V. and Pérez-Richet, E. (2022). Test design under falsification.Econometrica, 90(3):1109–1142

2022
[19]

Titova, M. (2022). Persuasion with verifiable information. Working paper

2022
[20]

Yang, L. L. (2024). Information design with costly state verification. Discussion Paper 502, CRC TR 224, University of Bonn and University of Cologne. 30

2024

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.

[1] [1]

and Babichenko, Y

Arieli, I. and Babichenko, Y. (2019). Private Bayesian persuasion.Journal of Economic Theory, 182:185–217

work page 2019

[2] [2]

Ben-Porath, E., Dekel, E., and Lipman, B. L. (2019). Mechanisms with evidence: Commitment and robustness.Econometrica, 87(2):529–566

work page 2019

[3] [3]

and Morris, S

Bergemann, D. and Morris, S. (2016). Information design, bayesian persuasion, and Bayes correlated equilibrium.American Economic Review: Papers & Proceedings, 106(5):586–591

work page 2016

[4] [4]

and Tirole, J

Caillaud, B. and Tirole, J. (2007). Consensus building: How to persuade a group.American Economic Review, 97(5):1877–1900

work page 2007

[5] [5]

and Skreta, V

Doval, L. and Skreta, V. (2022). Mechanism design with limited commitment.Econometrica, 90(4):1463–1500

work page 2022

[6] [6]

Gehlbach, S., Luo, Z., Shirikov, A., and Vorobyev, D. (2021). A model of censorship, propaganda, and repression. Working paper

work page 2021

[7] [7]

Grossman, S. J. (1981). The informational role of warranties and private disclosure about product quality.The Journal of Law and Economics, 24(3):461–483

work page 1981

[8] [8]

and Treisman, D

Guriev, S. and Treisman, D. (2019). Informational autocrats.Journal of Economic Perspec- tives, 33(4):100–127

work page 2019