{"id":"cff9f5b8-8eec-4727-af67-37f8b0db47cc","arxiv_id":"2411.15677","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":7,"one_line_summary":"In a simulated zero-sum media competition with bounded-rational players, equilibrium strategies reproduce the pattern that hyper-partisan sources spread more misinformation than centrists and that one side's misinformation triggers the other's.","lead":"This paper models two rival teams of news outlets as players in a zero-sum game where spreading misinformation buys attention but erodes credibility, and finds that equilibrium strategies push hyper-partisan outlets to misinform while centrists stay credible. The result is a formal account of how media competition can create an arms race in misinformation, and it suggests which policy levers might reduce polarization.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central results are driven by the assumed radicalization objective in Eq (6); the paper has not shown that competition itself, rather than this objective, produces hyper-partisan misinformation and the arms race.","rationale":"The reader's verdict is CONDITIONAL, and my stress test supports that assessment rather than overturning it. The paper is internally coherent: Eq (3) correctly gives misinformation a short-term influence advantage through the decay exponent, Eq (1) implements the credibility cost, and the numerical experiments are extensive. The most load-bearing weakness is the payoff function in Eq (6), which assumes that news sources maximize ideological radicalization and treat centrists as undesirable. That assumption is not derived from evidence about media firms' actual incentives; it is introduced as a modeling choice in the Equilibrium Concept section. Since the central qualitative claims—hyper-partisan sources choosing misinformation and the opponent following suit—are equilibrium consequences of this payoff, the title-level conclusion is conditional on the assumed objective. The proposed computational test would show whether the result is robust to a more standard audience- or profit-oriented objective. If the pattern disappears under such alternatives, the paper's contribution should be reframed as a model of radicalization-seeking outlets rather than a general account of media competition. This does not require changing the conditional verdict; it strengthens the conditions under which the paper should be accepted, namely a robustness analysis of the payoff specification and a quantitative comparison to the Ad Fontes data.","tokens_in":17635,"tokens_out":9434,"duration_ms":93035,"concrete_test":"Re-run the matrix-game equilibrium of the 'Limited Action' section (Eqs. 9-10) with the same parameters (λ, η, ξ, β, τ) and the same 9 action profiles, but replace Eq (6) with an empirically motivated audience/reach reward, e.g., r(x) = Σ_i (1 - |x_i|) or a reward proportional to the engagement-weighted attention implied by Eq (3). If the qualitative predictions—hyper-partisan sources sharing more misinformation and a best-response arms race—disappear or reverse, then the central claim rests on the radicalization objective rather than on competition. If they persist across reasonable alternative objectives, the concern is substantially resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Eq (6) defines r(xt,ct) = -Σ_i sin((π/2)x_i^t)^5, so both players' utility rewards moving public opinion to the ideological extremes and actively penalizes centrists and swing voters. This radicalization-seeking objective is the only reason misinformation is attractive: misinformation boosts short-term influence in Eq (3) but erodes credibility via Eq (1), and that erosion only helps a player if the goal is to concentrate opinion at one's own pole while depriving the other side of the center. The paper does not derive this utility from any observed media-firm objective, such as audience size, advertising revenue, credibility maximization, or shareholder value; it is asserted in the 'Equilibrium Concept' section. The zero-sum formulation in Eq (5) further builds in the assumption that one side's gain is exactly the other's loss. Because the headline claims are about competition between news sources, the argument needs the conditional 'if outlets maximize ideological radicalization, then...' to be stated and defended. Without a robustness check against alternative, empirically grounded objectives, the model's equilibrium pattern and the misinformation arms race are not evidence about media competition generally. This is a missing justification for a conclusion that is structurally encoded in the payoff, not an internal inconsistency.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes a computational game-theoretic model of news-source competition. Two coalitions ('left' and 'right') choose, at each time step, whether each source under their control disseminates factual news or misinformation. Source credibility evolves according to Eq. (1), individual opinions by the stochastic social-media interaction in Eq. (2), and media influence by Eq. (3). The competition is formulated as a zero-sum sequential game (Eq. 5) with the running reward in Eq. (6), which rewards moving public opinion to the ideological extremes. The authors solve for quantal response equilibria numerically, first with a complete action space and then with a restricted set of strategy profiles (Fig. 4e), and report that equilibrium play reproduces the Ad Fontes-style credibility-bias relation, that misinformation polarizes opinion and creates echo chambers, that one player's increased misinformation leads the opponent to increase misinformation as well, and that sensitivity analysis identifies possible interventions.","tokens_in":17956,"tokens_out":4370,"duration_ms":42819,"significance":"Conditional on the modeling assumptions, the paper offers an integrative framework and produces falsifiable comparative statics, such as phase transitions in the misinformation gain η and credibility gain ξ, and monotone effects of rationality τ. The use of quantal response equilibrium to model bounded rationality is appropriate for the problem, and the authors are transparent about the local-solution limitation of their function approximation. If the central mechanism were robust, the paper would be a useful proof of concept that competitive equilibria can rationalize hyper-partisan misinformation. However, the headline empirical claim is only qualitative, the core mechanism is substantially encoded in the payoff, and the numerical results are not accompanied by code, data, or convergence diagnostics. The significance as a statement about real media competition is therefore not yet established.","major_comments":[{"comment":"The reward function r(xt,ct) = -Σ_i sin((π/2)x_i^t)^5 assumes that media players seek to radicalize public opinion: it is maximized at the extremes and minimized at the center, and the text explicitly says swing voters are undesirable and ϑ=5 is chosen to make centrist opinions low-reward. This radicalization-seeking objective is asserted rather than derived from observed media-firm goals such as audience size, advertising revenue, or credibility maximization. Since Eq. (5) is zero-sum and Eq. (6) is the only payoff, the results that equilibrium play involves misinformation, polarization, and an arms race are largely consequences of this assumed objective. The central claim that competition fuels misinformation therefore requires either an empirical defense of Eq. (6) or a robustness analysis with alternative, empirically grounded objectives.","section":"Equilibrium Concept, Eq. (6)"},{"comment":"The claimed reproduction of the real-world credibility-bias distribution is qualitative and partly built into the action set. The complete-action equilibrium in Fig. 4(a) produces sharp credibility transitions, and the authors concede that this result is 'contrary to the real-world credibility distributions'; they then introduce the limited action set in Fig. 4(e). The nine profiles in Fig. 4(e) are hand-selected strategy profiles that already have a smooth credibility-bias shape, so the equilibrium selecting among them cannot be presented as independent evidence that the model reproduces Fig. 1(a). No goodness-of-fit statistic, error bar, or comparison to a null model is provided. As it stands, the abstract's claim that 'the resulting equilibria for this game reproduce the credibility-bias distribution' is not supported by the evidence.","section":"Results, Fig. 4"},{"comment":"The game in Eq. (5) is symmetric between the two players: the payoff depends only on the opinion vector and not on which player controls which sources. A symmetric zero-sum game cannot endogenously generate a systematic difference in misinformation policies between the two parties without an explicit symmetry-breaking mechanism such as asymmetric initial conditions, asymmetric susceptibility, or asymmetric action sets. The paper nevertheless suggests that its findings explain the partisan imbalance in misinformation exposure documented by Mosleh and Rand. This inference is unsupported without such a mechanism or a clear statement of what breaks the symmetry in the simulations.","section":"Results, Fig. 5"},{"comment":"The central equilibrium results are numerical but not reproducible from the manuscript. The value function is approximated with function approximation, the payoff matrix is estimated empirically through 200 simulations, and the extragradient method is used; however, the paper provides no code, data, hyperparameter ranges, random seeds, or convergence diagnostics. Figure 7 further states that increasing τ makes the quantal response equilibrium computation unstable, and the text notes that function approximation methods typically find local solutions. The assertion that the reported equilibrium is stable is therefore not verifiable from the material supplied. Given that the paper's main conclusions rest on these computations, this is a load-bearing gap.","section":"Quantal Response Equilibrium and Results"}],"minor_comments":[{"comment":"The parameter for credibility gain is written as ζ in Eq. (3) but as ξ in the surrounding text and in later sensitivity analysis; this inconsistency should be fixed throughout.","section":"Eq. (3)"},{"comment":"The reward is written as r(xt,ct), but the right-hand side depends only on xt, not on ct; either remove the unused argument or define the dependence.","section":"Eq. (6)"},{"comment":"The notation R(qt) is introduced without definition; it should be connected to the running reward defined in Eq. (6) or defined explicitly.","section":"Eq. (8)"},{"comment":"The text says that player R controls sources with ym < 0, which duplicates the definition of player L; the second occurrence should presumably read ym > 0.","section":"Information Structure and Equilibrium Concept"},{"comment":"There are several typos and stylistic errors, including 'influnced', 'also remains also', and inconsistent subfigure references in the Figure 2 caption; a careful proofread is needed.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this paper deserves a serious referee, but the referee should push for major revision. The genuinely new piece is the integration: a credibility-susceptibility opinion dynamics model (Eqs 1–3) coupled to a zero-sum quantal response equilibrium over media strategies, plus the arms-race comparative static and the intervention analysis. That combination is not in the cited literature, and the paper is transparent about what it does and does not do. The sensitivity analysis around misinformation gain, credibility gain, rationality, and susceptibility is exactly the kind of exploration that makes a modeling paper useful.\n\nThe soft spots are real, and I largely agree with the stress-test. Eq (6) is load-bearing and asserted, not derived. The reward −Σ sin((π/2)x_i)^5 means both players want to radicalize the public and actively dislike swing voters. Under that objective, misinformation is attractive despite the credibility penalty. That gives a coherent conditional result — if media outlets maximize ideological radicalization, then competition pushes them toward misinformation — but the paper keeps stating the unconditional version. The zero-sum assumption in Eq (5) further builds in the adversarial frame. Without robustness checks against alternative objectives like audience reach, advertising revenue, or credibility maximization, the headline claim about competition is not established.\n\nSecond, the empirical validation is qualitative. The match to the Ad Fontes credibility-bias curve is visual. The complete-action equilibrium does not reproduce the smooth curve; the paper says so and introduces a hand-picked limited action set (Fig 4e) that does. That is an honest limitation, but it weakens the claim that the model “reproduces” real-world distributions. There is no code, no data, no convergence diagnostics for the QRE solver, and several parameters appear only inside the text. The stability statement is also based on a local function approximation.\n\nCredit where due: the paper cites the right adjacent literatures — strategic disinformation games, opinion dynamics, media competition — and the self-citations are not padding. The writing is clear, and the authors flag the main modeling simplifications themselves.\n\nFor whom: computational social scientists and misinformation policy people will get value from the model and the sensitivity results, but they should treat the equilibrium predictions as hypotheses. Recommendation: accept for peer review, require major revision. The referee should ask for a derivation or empirical grounding of Eq (6), robustness to alternative objectives, quantitative fit statistics, and code/data release.","headline":"A novel and honest model of media competition and misinformation, but its central empirical claim rests on an assumed radicalization objective and a qualitative fit — worth serious peer review, with major revision required.","tokens_in":18446,"tokens_out":1981,"would_cite":false,"duration_ms":20429,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper argues that competition between ideologically opposed groups of news sources rationally produces misinformation-heavy strategies for hyper-partisan outlets, public polarization, and a misinformation arms race.","keywords":["misinformation","media competition","opinion dynamics","quantal response equilibrium","zero-sum game","source credibility","polarization","misinformation exposure"],"falsifier":"Replace the reward in Eq. (6) with an audience-reach proxy, such as the total number of influenced individuals regardless of opinion, keep every other equation and parameter fixed, and recompute the quantal response equilibrium; if hyper-partisan sources no longer disseminate the most misinformation, the paper's explanation of real-world misinformation rests entirely on the radicalization objective rather than on competition itself.","tokens_in":17478,"feed_emoji":"📰","tokens_out":12025,"duration_ms":98102,"temperature":0.7,"pith_summary":"News outlets that compete for public influence face a dilemma: sharing misinformation boosts immediate attention but erodes their credibility. This paper models that trade-off as a zero-sum sequential game between two coalitions of ideologically opposed sources, solved with the quantal response equilibrium, a standard account of boundedly rational decision-making. It claims that in equilibrium, hyper-partisan outlets rationally disseminate more misinformation than centrist ones, reproducing the credibility-bias pattern seen in real-world media ratings. It further claims that competition creates a misinformation arms race—when one side increases misinformation, the opposing side's best response is to do the same—and that these dynamics polarize public opinion. The payoff is a strategic explanation for why misinformation persists despite its reputational costs, together with levers for intervention.","feed_headline":"Media competition pushes news outlets to spread misinformation","feed_subtitle":"Model reproduces real-world trust gaps and predicts a fake-news arms race.","key_machinery":"The central object is a zero-sum sequential game coupling two dynamic processes. Source credibility evolves as a convex combination of past credibility and the current factual-or-misinformation action, $c_{t+1} = \\lambda c_t + (1-\\lambda) a_t$, and individual opinions evolve under social influence (an exponential-decay homophily kernel) plus media influence, whose multiplier $\\psi(r,c,a,s) = \\exp[-\\hat{\\kappa}(1+\\eta a)(1+\\zeta(1-c)(1-s)) r]$ encodes the attention gain from misinformation ($\\eta$) and the credibility penalty ($\\zeta$). The solution concept is the quantal response equilibrium of this entropy-regularized game, which captures bounded rationality; the equilibria are computed with the extragradient method on an empirically estimated payoff matrix. The running reward $r(x_t,c_t) = -\\sum_i \\sin(\\frac{\\pi}{2} x_i^t)^5$ is what makes the game zero-sum and encodes the assumption that players seek to radicalize public opinion toward their own pole while treating swing voters as undesirable.","core_discovery":"The paper's central discovery is that competition for public influence between two ideologically opposed coalitions of news sources has quantal-response equilibria in which the optimal strategy for hyper-partisan sources is to disseminate misinformation often, while centrist sources preserve credibility by sharing mostly factual content. This equilibrium pattern matches the credibility-by-bias distribution observed in real-world news-source ratings without any curve fitting: it emerges from the structural incentives of the game. The same equilibria show that misinformation polarizes the opinion distribution into echo chambers, and that deviations by one player—such as increasing misinformation output—are met by the opposing player's optimal response of also increasing misinformation, producing a reciprocal arms race. The model also predicts a phase transition: if the short-term attention gain from misinformation is reduced enough, or the long-term credibility penalty increased, the equilibrium flips, depolarizing the community but sharply raising average misinformation exposure because centrist sources then become the main misinformation spreaders.","pith_inferences":["If the radicalization reward were replaced by a profit- or reach-maximizing objective, the qualitative results might change substantially; comparing those equilibria would reveal how much of the misinformation outcome is driven by the assumed payoff versus the structure of competition.","The arms-race prediction implies a testable natural experiment: if one outlet's misinformation output exogenously increases (for example, after a credibility-rating event), competitors' misinformation rates should subsequently rise; if they do not, the equilibrium response mechanism is incomplete.","The model assumes susceptibility is fixed, but psychological evidence of an illusory-truth effect suggests exposure to misinformation can itself raise susceptibility; incorporating that feedback would likely strengthen the self-reinforcing spread and could shift the phase-transition threshold."],"forward_implications":["Reducing the short-term engagement payoff of misinformation—for example by debunking from within the same partisan community—is predicted to be more effective at depolarizing public opinion and lowering exposure than merely increasing penalties on source credibility.","A single low-credibility source can disrupt the equilibrium and trigger a system-wide escalation: the opposing coalition's optimal response is to increase its own misinformation output, producing a lose-lose outcome for both sides.","Improving community media literacy (lowering susceptibility) tends to depolarize, but pushing susceptibility too low can flip the equilibrium into a phase where centrist sources become the main misinformation spreaders and average exposure spikes.","As players become more rational (approaching Nash equilibrium), the equilibrium becomes more polarized, implying that misinformation-heavy hyper-partisan strategies are calculated responses to the competitive landscape rather than mistakes."],"supporting_citations":[{"why":"Cites research that mainstream media are major propagators of fake news, motivating the focus on media competition.","marker":"4"},{"why":"Supports the premise that fake news has intrinsic attention value, which the model encodes as misinformation gain η.","marker":"5"},{"why":"Empirical or strategic result that disinformation outperforms honesty in competition for social influence, the basis of the gain-vs-credibility trade-off.","marker":"26"},{"why":"Provides real-world exposure data on social media used to validate the model's predicted misinformation-exposure distribution.","marker":"34"},{"why":"Supplies the Lagrange-multiplier technique for incorporating rationality constraints into the zero-sum game.","marker":"42"},{"why":"Prior game-theoretic model of online misinformation whose equilibrium analysis this paper extends to competing news coalitions.","marker":"43"},{"why":"Supplies the exponential-decay homophily kernel used for social and media influence in the opinion dynamics.","marker":"62"},{"why":"Defines quantal response equilibrium for normal-form games, the bounded-rationality solution concept used throughout.","marker":"69"},{"why":"Extends quantal response equilibrium to extensive-form (sequential) games, matching the paper's dynamic game setup.","marker":"70"},{"why":"Provides the fast extragradient algorithm used to compute the equilibrium of the entropy-regularized matrix game.","marker":"72"}],"fun_headline_variants":["Misinformation arms race emerges from media competition","Media rivalry creates misinformation arms race, model shows","Competition pushes news outlets into misinformation arms race","How media competition triggers a fake-news arms race","News competition fuels misinformation and polarization"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is the reward function, which assumes media players maximize the ideological radicalization of the public (via $r(x_t,c_t)=-\\sum_i \\sin(\\frac{\\pi}{2} x_i^t)^5$) rather than audience reach, advertising revenue, or per-source credibility.","fun_headline_variants_meta":{"raw":{"variants":["Misinformation arms race emerges from media competition","Media rivalry creates misinformation arms race, model shows","Competition pushes news outlets into misinformation arms race","How media competition triggers a fake-news arms race","News competition fuels misinformation and polarization"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000178,"raw_usage":{"total_tokens":1294,"prompt_tokens":943,"completion_tokens":351,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":559,"completion_tokens_details":{"reasoning_tokens":283}},"tokens_in":559,"tokens_out":351,"duration_ms":3674,"temperature":1.0,"reasoning_tokens":283,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T14:01:32.465357+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Replace the reward in Eq. (6) with an audience-reach proxy, such as the total number of influenced individuals regardless of opinion, keep every other equation and parameter fixed, and recompute the quantal response equilibrium; if hyper-partisan sources no longer disseminate the most misinformation, the paper's explanation of real-world misinformation rests entirely on the radicalization objective rather than on competition itself.","supporting_citations":[{"cited_title":"Causes and consequences of mainstream media dissemination of fake news: literature review and synthesis","cited_arxiv_id":null,"evidence_quote":"Cites research that mainstream media are major propagators of fake news, motivating the focus on media competition."},{"cited_title":"& Brassard, G","cited_arxiv_id":null,"evidence_quote":"Supports the premise that fake news has intrinsic attention value, which the model encodes as misinformation gain η."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Empirical or strategic result that disinformation outperforms honesty in competition for social influence, the basis of the gain-vs-credibility trade-off."},{"cited_title":"& Rand, D","cited_arxiv_id":null,"evidence_quote":"Provides real-world exposure data on social media used to validate the model's predicted misinformation-exposure distribution."},{"cited_title":"Balancing Two-Player Stochastic Games with Soft Q-Learning","cited_arxiv_id":"1802.03216","evidence_quote":"Supplies the Lagrange-multiplier technique for incorporating rationality constraints into the zero-sum game."},{"cited_title":"& Siderius, J","cited_arxiv_id":null,"evidence_quote":"Prior game-theoretic model of online misinformation whose equilibrium analysis this paper extends to competing news coalitions."},{"cited_title":"& Helbing, D","cited_arxiv_id":null,"evidence_quote":"Supplies the exponential-decay homophily kernel used for social and media influence in the opinion dynamics."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines quantal response equilibrium for normal-form games, the bounded-rationality solution concept used throughout."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Extends quantal response equilibrium to extensive-form (sequential) games, matching the paper's dynamic game setup."},{"cited_title":"& Chi, Y","cited_arxiv_id":null,"evidence_quote":"Provides the fast extragradient algorithm used to compute the equilibrium of the entropy-regularized matrix game."}],"review_version":1}