| paper | d4-conf-ecai-2010-0066 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | yes |
Theorem 3 establishes convergence as the number of agents grows, but makes no computational claim. Continuous Approval Threshold is a well-posed problem, and the proposed SAT reduction gives it a complexity result, but that result is new analysis rather than a computational result of this paper.
fails bit a — no named computational result to mirror
The proposed problem covers finding a world with maximum approval mass; it leaves the paper’s probabilistic convergence theorem, Truth Tracking postulate, operator classifications, and experiments outside the mirror.
There is a plausible continuous-population question here, but the paper has no named computational-complexity result to serve as an anchor under your rules. Theorem 1 is Condorcet’s cited two-world limit theorem; Theorem 2 is the cited List–Goodin limit theorem; and Proposition 1, Proposition 2, Proposition 3, and Proposition 4 concern probabilistic scores, reliability notions, or truth-tracking properties of operators. None asserts that a computational problem is in P, NP-hard, FPT, or similar.
The closest substantive result is Theorem 3, proved in this paper. It says that for a finite set of worlds and a profile of \(n\) independent, homogeneous, \(R4\)-reliable agents, each of whom may report a base with several models, the probability that the true world’s approval score exceeds any particular other world’s score tends to \(1\) as \(n\) grows. Here \(R4\)-reliability means that each agent includes the true world in their base with greater probability than they include any other world. This is an analytic limit result, not a computational one.
A reasonable mirror of the approval-merging problem studied around Theorem 3 would be Continuous Approval Threshold. An instance consists of a finite set \(P\) of propositional variables, a support of bases \(K_1,\ldots,K_\tau\), binary-rational masses \(\mu_1,\ldots,\mu_\tau\geq 0\) summing to \(1\), and a rational threshold \(\theta\). A base is a type; \(\mu_i\) is the fraction of the population with that base. For each world \(\omega\in\{0,1\}^{|P|}\), its approval mass is
\[
s_\mu(\omega)=\sum_{i:\,\omega\models K_i}\mu_i.
\]
The question is whether some world has \(s_\mu(\omega)\geq\theta\); a solution is such a world, or “no” if none exists. Varying the threshold recovers the optimization problem of finding approval winners.
This has a credible high-multiplicity setting: a large fact-checking or survey panel uses a small collection of recurring evidence and reporting templates. Agents with the same base are indistinguishable to the approval rule, so \(N\) can be large while \(\tau\ll N\). In a truth-tracking version, the reporting protocol and its conditional behavior would also be part of an agent’s complete type; independence and \(R4\)-reliability would describe the population generating the reports.
For this formulation I expect Class B hardness. With \(q\) clauses of a 3-CNF formula, use each clause as a base with mass \(1/q\), and set \(\theta=1\). A world reaches that threshold exactly when it satisfies every clause, so the threshold problem is NP-hard (and is in NP). Replicating the profile preserves its masses and support, giving realizations with arbitrarily many agents per type. The hardness comes from the propositional agenda, not from population size. This is a new complexity question motivated by the paper’s approval rule; it is not a complexity result proved in the paper.
The mirror is plausible because it turns the paper’s approval scores from counts into population fractions and asks for the computational cost of computing the resulting merge. Theorem 3 supplies a useful reason to care about that output: under its sampling assumptions, approval selects the true world with probability tending to \(1\). The mirror covers that approval-based part of the paper, not the full Truth Tracking postulate, every merging operator, or the convergence experiments.
The weak point is also decisive for the requested rubric: the paper’s limit theorem does not establish the complexity of the approval problem. Thus this is a concrete, potentially author-recognizable continuous mirror, but it cannot be presented as an anchor-compliant case based on the paper’s named computational results.
The strongest case against a mirror is that this paper has no named computational result to continuize. Theorem 3 is its main new result, but it is a limit theorem about the probability that approval scores rank the true world highest. Theorem 1 and Theorem 2 are cited results; the postulate and operator results classify logical or epistemic properties. None asks for an algorithm or establishes a complexity bound. That is a gap in the proposed anchor, not a complaint that the paper’s results change in the limit.
The proposed Continuous Approval Threshold problem is well-defined, and the clause construction does show it can be hard. But it is a new optimization problem attached to the approval rule, not a computational question Theorem 3 raises. Its SAT reduction shows that the chosen formulation can encode satisfiability; it says nothing about the theorem’s convergence claim. Giving each base a rational mass makes the input a continuous society in the programme’s sense, but does not supply the missing computational result.
The better direct mirror would take an exact mass distribution over bases and ask for the world with maximum approval mass. Under the theorem’s \(R4\) condition, the true world has strictly greater expected approval mass than each competitor. The continuum therefore removes the sampling fluctuation that gives Theorem 3 its content. Computing the maximizing world remains a possible computational question, but it is a separate analysis of approval merging, not a continuization of the theorem’s probabilistic result.
One can preserve the theorem’s sampling content by asking for the probability that a finite sample of \(n\) agents yields the true world as approval winner. That would connect directly to the experiments, but the computational object would be a finite-sample probability calculation over a reporting distribution. It is a plausible follow-on question; it does not turn this paper into a computational study of a continuous society.
This does not show that no worthwhile continuous approval problem could be posed. A large panel with recurring belief-base templates is a sensible high-multiplicity setting, and the exact weighted-winner problem may be worth studying on its own. The negative case is narrower: the paper supplies no computational result to anchor that study, and Theorem 3’s contribution—convergence under reliability and independence—either becomes an immediate expected-score comparison in the exact continuum or remains a finite-sample probability question. The case against the universal claim is weak precisely because that weighted-winner problem could still be a worthwhile new mirror if its authors recognize it as their problem.
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.