| paper | d4-journals-geb-2017-0130 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | no |
Theorem 2 constructively designs an incentive-compatible mechanism, and Theorem 1 explicitly constructs an equivalent rule, so I treat the named results as computational constructions even without complexity bounds. But in an atomless population, one agent has no effect on the lottery chosen from aggregate reports, making the paper's incentive constraint vacuous; retaining individual influence requires finite-population pivotality and changes the question.
fails bit b-changes — the continuum limit changes the question
The candidate would concern Theorem 1's ordinal-rule implementation and Theorem 2's superior cardinal rule; it would leave Proposition 2's scoring-rule connection and Section 7's numerical findings without a continuous computational counterpart.
The strongest honest positive case is that this paper suggests a plausible high-multiplicity *follow-up*, but it has no named computational result of the kind this task requires as an anchor.
The closest result is Theorem 2, proved in the paper: in its neutral Bayesian environment with at least three alternatives, for \(n\ge5\) there exists an incentive-compatible cardinal voting rule whose utilitarian social welfare is higher than that of every ordinal rule. The paper also proves Theorem 1: every ordinally Paretian ex-post rule has an incentive-compatible ordinal rule that gives each agent the same expected utility. Both are existence and welfare results, not statements that an underlying problem is in P, hard, or otherwise computationally classified. The numerical LP in Section 7 is not a numbered computational result and gives no complexity bound.
A plausible regime for a follow-up would be a large representative-democracy electorate with many citizens but relatively few recurring value profiles. A type would need to be each citizen’s full normalized cardinal value vector, since rankings alone discard the intensities central to the paper; \(\mu_t\) would be the fraction of the population with vector \(t\). Theorem 2’s split of each ranking type into high- and low-intensity subtypes gives a real structural reason to explore such a model. But that does not make Theorem 2 a computational anchor.
The main obstacle is preserving the paper’s incentive question. If a mechanism chooses a lottery from the reported population distribution, one person has zero mass in the continuum and cannot change that distribution. Their incentive constraint is then vacuous. Keeping the effect of one report would require a finite-population or first-order pivotality formulation, which is a new model and is not established by the paper’s named results. I would expect the authors to recognize it as a possible asymptotic extension, but not as the continuous form of a computational problem they solved.
So there is no valid per-anchor prover question to state here. The population scenario is plausible, and the paper’s intensity-based mechanism design offers motivation, but presenting it as an anchored continuous computational mirror would overstate the paper’s results.
The strongest objection comes before the limit: the paper names no computational result to mirror. Theorem 1 and Theorem 2 are existence and welfare comparisons; Proposition 2 identifies a rule class. Section 7 solves a discretized LP for a fixed two-agent, three-alternative example, but gives no complexity classification. A complexity question about finding an optimal incentive-compatible mechanism would be a new problem, not a computational result of this paper.
Theorem 2 is the best candidate for a mirror, and its incentive constraints expose the main difficulty. In the natural large-electorate model, let the society be a distribution \(\mu\) over complete cardinal value vectors, and let an anonymous rule choose a lottery from that distribution. A single agent’s report changes \(\mu\) by zero, so it cannot change the lottery. Every report is then equally good for that agent: Bayesian incentive compatibility, the feature that makes Theorem 2 interesting, has become vacuous. Keeping a deviation of size \(1/n\) preserves the finite agent’s influence, but the constraints then depend on a finite population scale. A rescaled pivotality condition might be worth studying, but the paper does not define one, and it would be a new model.
The stronger proposed follow-up—compute the welfare-maximizing IC mechanism for a large representative-democracy electorate with recurring value profiles—runs into the same fork. If exact values have finite support, repeated full types are possible, but preserving an individual’s effect means retaining the finite change in the count of reports. That is a high-multiplicity finite-agent problem, not the atomless population limit. If the population is genuinely atomless, individual incentives disappear. Grouping the paper’s high- and low-intensity classes does not avoid this: agents in each class can still have different cardinal utilities, so those classes are not complete types in the programme’s sense.
The paper’s iid continuous value model also has almost surely no repeated complete value vectors. That alone does not rule out a finite-support regime, but the finite-support version still faces the incentive problem above. Theorem 1 fares no better: in an atomless aggregate-report model, incentive compatibility becomes automatic rather than a computational question about the ordinal rules it characterizes.
I cannot rule out a worthwhile, separately motivated complexity theory for mass mechanisms with pivotality refinements. The case against is that this paper supplies neither a computational anchor nor a non-vacuous population-continuous version of its central incentive question. A worthwhile extension may exist; the material here does not make it a continuous computational mirror.
The adversarial triple: the proponent anchors on up to three named results; the opponent sees that case and must defeat every anchor; the judge decides which case convinced it. These are the pipeline’s own outputs, generated by tools/triple_run.py — no human edited them. The paper’s own text is not reproduced here beyond the quoted statement above.