| paper | Deliberation and Voting in Approval-Based Multi-Winner Elections |
| authors | Kanav Mehra, Nanda Kishore Sreenivas, Kate Larson |
| venue | IJCAI 2023 |
| filed under | multiwinner · multiwinner |
| judged by | gpt-5.6-luna / xhigh (triple__luna__xhigh__c2r1) |
| judge confidence | high |
| authors would recognise it | yes |
Red because the paper contains no numbered theorem, lemma, corollary, or proposition asserting a computational result; its PAV hardness and AV polynomiality claims are cited or informal background. The high-multiplicity PAV winner problem is a sensible direct mirror of a rule used in the paper, but it is not a result of this paper and cannot satisfy bit (a). The deliberation dynamics also require schedules or histories, so they do not provide a clean anchored computational question.
fails bit a — no named computational result to mirror
The latent mirror covers PAV winner selection over repeated approval-ballot types. It leaves deliberation, group formation, welfare and representation ratios, EJR/PJR observations, and the simulation comparisons unmirrored.
The strongest honest answer is that this paper has no admissible computational anchor. In the supplied text, there is no numbered Theorem, Lemma, Corollary, or Proposition asserting a complexity or algorithmic result. The closest claims are the unnumbered statement that “computing PAV is NP-hard” in Section 2.2, explicitly cited to Aziz et al. (2014), and the informal statement that AV is polynomial in Section 6. Neither is a named result of this paper.
So, strictly under your rules, there are zero anchors and hence no valid “one continuous problem per anchor.” Any stronger case would have to invent a theorem number or treat a cited background fact as a result proved by Mehra, Sreenivas, and Larson.
The closest latent mirror would be the following, though it cannot serve as a compliant anchor. Given candidates \(C\), committee size \(k\), a finite set \(T\) of approval-ballot types, and rational masses \(\mu_t\), choose \(W\subseteq C\), \(|W|=k\), maximizing
\[ \operatorname{PAV}_{\mu}(W) = \sum_{t\in T}\mu_t h(|A_t\cap W|), \qquad h(r)=\sum_{j=1}^{r}\frac{1}{j}. \]
This is a natural high-multiplicity version of the PAV winner problem. A plausible regime is a large participatory-budgeting electorate containing many repeated civic profiles—citizens sharing the same approval ballot, and, if welfare is also modelled, the same utility vector—so \(|N|\gg |T|\). The expected classification is Class B: the discrete problem embeds directly, and the hard combinatorics lies in selecting candidates for the committee rather than in population multiplicity.
That mirror is mathematically sensible, but it covers only the paper’s background discussion of PAV and not its central contribution: deliberation followed by voting. A continuous deliberation model could instead let types include \(A\), utility vector \(u\), group identity, confidence \(\Delta\), and influence parameters, with mass assigned to type-level deliberation groups and the bounded-confidence update applied to those groups. That would be a worthwhile new computational-social-choice problem, but it is not anchored in any named computational result here.
The fatal weakness is therefore bibliographic and structural, not a defect in continuization: this is an experimental paper whose formal results are definitions, simulations, and observations. It may inspire a good continuous mirror, but it does not itself provide the required named theorem to anchor one.
The strongest negative case is bibliographic and structural: this paper does not supply an admissible computational anchor.
There is no numbered Theorem, Lemma, Corollary, or Proposition establishing a complexity or algorithmic result. “Computing PAV is NP-hard” is cited background from Aziz et al., not a result proved by this paper; “AV is polynomial” is likewise informal context. The actual contributions are simulation outcomes, statistical comparisons, and empirical observations about deliberation. None defines a worst-case computational problem whose continuous counterpart the ChoCo programme could classify.
The proponent’s latent PAV formulation,
\[ \max_{W\subseteq C,\ |W|=k} \sum_{t\in T}\mu_t h(|A_t\cap W|), \]
is mathematically legitimate. Repeated approval-ballot types in a large participatory-budgeting electorate give a sensible high-multiplicity regime. I would not object that this is “already done,” nor would I object merely because the hardness lies in committee selection. But it is a mirror of the cited PAV literature, not of a result of Mehra, Sreenivas, and Larson. It also omits the paper’s central contribution: deliberation. Coupling PAV to deliberation would be a new problem, not a continuous reformulation of anything computationally established here.
The stronger rescue is to let a type contain \(A\), \(u\), group identity, \(\Delta\), \(\alpha\), and \(\beta\), assign masses to such types, and define type-level bounded-confidence dynamics. That is conceivable, but the paper’s dynamics are not determined by type masses alone. Outcomes depend on the realized speaker order, group partition, and repeated pair exposures. Two populations with identical type-mass vectors can therefore produce different final preferences. Preserving that information requires an interaction schedule, a coupling over groups, or a distribution over histories. Making the schedule part of the input restores the individual interaction structure; averaging over random schedules produces a new mean-field model; optimizing schedules produces a new group-formation problem. Each could be worthwhile research, but none is a computational result of this paper.
The paper’s actual experimental regime also samples utilities and deliberation parameters independently from continuous distributions, so complete repeated types occur with probability zero. A repeated-cohort version can certainly be invented, but that is a counterfactual model rather than the high-multiplicity form of the studied experiments.
Thus the negative verdict should be narrow but firm: this paper should not enter ChoCo as a source of a continuous mirror. Its PAV mirror is valid but belongs to prior PAV work, while its deliberation mirror is an interesting new mean-field project with no named computational theorem to anchor it. A universal claim that no worthwhile continuization could ever be inspired by the paper would be too strong; the defensible claim is that the paper itself provides no admissible computational anchor.
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.