| paper | d4-conf-atal-2007-0251 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | yes |
The paper compares truth-tracking probabilities for finite electorates, but contains no numbered result asserting a computational claim. The proposed continuum limit is a recognizable statistical question, yet its fixed four-type example takes only constant enumeration and does not supply a computational mirror of a named result; bit (a) therefore fails.
fails bit a — no named computational result to mirror
The proposed limit covers at most the asymptotic truth-tracking comparison for the paper’s doctrinal example; it leaves the finite-\(N\) sampling comparisons and proposed generalizations to other agendas untouched.
There is no eligible anchor in this paper: it contains no numbered theorem, lemma, corollary, or proposition asserting a computational result such as tractability or hardness. The closest results are the unnumbered comparisons in Section 4: for the plotted finite electorates, fusion’s probability of selecting the correct judgment is better at small and middling competence levels. Those comparisons are not named complexity results, so I cannot honestly present a continuous problem as anchored to one.
The strongest plausible mirror is an unanchored limit of that truth-tracking comparison. Consider a large deliberative body judging \(P,Q,R\) under \((P\land Q)\leftrightarrow R\). Let the four consistent judgment types be \(T=\{S_1,S_2,S_3,S_4\}\), with mass meaning the fraction of members holding each type. In a regime where members share competence \(p\), and \(P\) and \(Q\) are independently true with prior probability \(q\), the continuum mass conditional on true judgment \(\sigma\) is
\[
\mu_t^{\sigma,p}
=\prod_{A\in\{P,Q\}}
p^{\mathbf 1[t(A)=\sigma(A)]}
(1-p)^{\mathbf 1[t(A)\ne\sigma(A)]}.
\]
For fusion, the selected judgments are
\[
M_F(\mu)=\arg\min_{\omega\models IC}\sum_{t\in T}\mu_t\,d_H(t,\omega).
\]
A precise candidate question is: given rational \(p,q\), does fusion’s limiting probability of selecting the correct judgment exceed that of both PBP and CBP? The probability averages over the paper’s prior on the four possible truths, and a tie counts as selecting the truth. This has a credible high-multiplicity setting—many members, at most four judgment types—and the fixed version is tractable by enumerating those types and outcomes.
Its weakest point is substantial: the paper’s headline comparison concerns finite-group sampling variation, while this limit replaces random group counts with deterministic masses. The middling-competence advantage may therefore change or disappear. And because the paper supplies no named computational anchor, this is only a plausible continuous analogue of its statistical question, not a supported computational-complexity mirror.
There is no named computational result here to mirror. The paper’s results are probabilistic comparisons of truth-tracking performance, chiefly for one three-proposition example; it does not state a complexity theorem for finding a fusion outcome or evaluating those probabilities. That is the strongest ground against a mirror in this paper.
The proposed limit is a sensible population model, but its fixed example does not create a computational problem. Conditional on truth \(S_1\), the four type masses are \(p^2,p(1-p),p(1-p),(1-p)^2\). Fusion compares four linear costs; for example, the cost of choosing \(S_1\) is \(3-2p-p^2\), while the costs of choosing \(S_2\) and \(S_3\) are both \(1+p^2\). Repeating this for the other three possible truths and averaging with prior \(q\) is a fixed amount of arithmetic. The mass vector is derived from \(p\) and the truth state, rather than supplied as a computational instance. This is a legitimate asymptotic truth-tracking calculation, but it is not the kind of complexity question the programme is aimed at.
The stronger candidate would make the agenda and integrity constraints variable, and ask for the complexity of evaluating truth-tracking performance or finding the fusion outcome from a distribution over judgment types. That could be an interesting computational problem. But the paper supplies no general model for how agents’ consistent judgments are generated conditional on truth beyond its specific doctrinal example. And asking for the minimizing model under arbitrary constraints shifts the subject from comparing procedures as truth-trackers to computing a weighted integrity-constrained fusion outcome. That is a new question, not a computational result of this paper with its population axis continuized.
I would not claim that the high-multiplicity regime is implausible: a large expert panel with four recurring judgment patterns is entirely credible, and the proposed limit has a clear meaning. The negative case is therefore strongest as a claim about fit: this paper offers a natural continuous statistical limit, but no worthwhile *computational mirror of its results*. Its weak point is that a broader agenda and a carefully specified truth-generation model might support a worthwhile ChoCo project; the paper itself does not give enough computational structure to make that case.
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.