| paper | Distances Between Top-Truncated Elections of Different Sizes |
| authors | Piotr Faliszewski, Jitka Mertlová, Pierre Nunn, Stanisław Szufa, Tomasz Wąs |
| venue | AAAI 2025 |
| filed under | voting · rationalization |
| judged by | gpt-5.6-luna / xhigh (triple__luna__xhigh__c2r1) |
| judge confidence | high |
| authors would recognise it | yes |
The paper contains no numbered theorem, lemma, corollary, or proposition asserting complexity, an algorithmic guarantee, or hardness; its named results are structural, axiomatic, or asymptotic. The proposed continuous evaluator is a sensible direct high-multiplicity formulation that the authors would likely recognise, but it is a new computational question rather than a mirror of named computational content from this paper. The mandatory computational-anchor gate therefore fails.
fails bit a — no named computational result to mirror
The proposed evaluator changes only the empirical voter average into a weighted average and introduces no population-dependent computational structure; consequently, it cannot supply the missing named computational anchor.
fatal: True
The mirror covers evaluation of the continuous positionwise distance and transfer of Theorem 3.4's UN-consistency; it leaves the swap-extension impossibility, DAP properties, empirical Kemeny asymptotics, and visualization results untreated.
The strongest honest conclusion is that this paper has no qualifying computational anchor under the ChoCo rule. Its numbered results are structural or axiomatic:
None asserts NP-hardness, polynomial-time solvability, parameterized complexity, approximation, or a related computational classification. The paper does say, unnumbered, that Wasserstein distance is computable in polynomial time, and that empirical \(i\)-Kemeny scores are generally intractable, citing Faliszewski et al. (2023). Neither is a qualifying named result of this paper. Thus, formally, this paper fails the named-anchor gate; that is not evidence that no continuous mirror exists.
The best salvage is a direct mirror of the positionwise-distance result, with Theorem 3.4 as a structural—not computational—anchor. Call it Continuous Positionwise Distance Evaluation\(_\infty\).
An instance consists of two candidate sets \(C,D\), with possibly different sizes, and two rational finite-support distributions \(\mu,\nu\) over complete top-truncated votes on \(C\) and \(D\). A type is exactly a top-truncated ballot: its strict ranked prefix and tied unranked remainder. Its mass is the fraction of voters casting that ballot. Define the continuous frequency matrices
\[ F_\mu=\sum_v \mu(v)\operatorname{freq}(v), \qquad F_\nu=\sum_u \nu(u)\operatorname{freq}(u). \]
If \(m=|C|\), \(r=|D|\), and \(s=\operatorname{lcm}(m,r)\), stretch the columns of \(F_\mu,F_\nu\) as in the paper and ask for
\[ \widehat d_{\mathrm{pos}}^\infty(\mu,\nu) = d_W\!\left(\operatorname{str}_s(F_\mu), \operatorname{str}_s(F_\nu)\right). \]
The solution is the exact distance, or equivalently the distance-threshold decision problem \(\widehat d_{\mathrm{pos}}^\infty(\mu,\nu)\le q\), together with an optimal matching of candidate columns. The objective and candidate matching are unchanged from the paper; only the voter multiset has become a population distribution.
This is a plausible high-multiplicity regime for large ranked-choice elections or constituency datasets with tens of thousands of voters, relatively few candidates, and a sparse collection of recurring top-truncated ballot types. Clearing denominators gives \(n_v/n\) masses and recovers exactly the finite election’s frequency matrix and distance. The authors should recognise this as their own construction with empirical averaging replaced by expectation, not as a different voting problem. It is population continuity only: candidates and rankings remain discrete.
The expected classification is Class A. The frequency matrices are computed by a linear scan over the distinct ballot types, and the remaining Wasserstein and candidate-assignment computations are polynomial. This is a genuine continuous computational question, although the polynomial-time claim would be a new ChoCo result, not a theorem already established in this paper. Theorem 3.4’s UN-consistency transfers directly; I would not claim the paper’s DAP results without a separate treatment of continuous medoid selection.
The mirror’s weakest point is decisive: the paper is primarily a distance-design and visualization paper, not a computational-complexity paper. A referee could reasonably regard Continuous Positionwise Distance Evaluation\(_\infty\) as a natural extension, but not as a continuization of one of the paper’s named computational results. Further questions would include whether a compact implicit distribution over ballot types preserves tractability, how rounding masses back to \(n\) voters affects the distance, and whether a continuous DAP functional has a useful exact complexity theory.
The decisive objection is the source gate. This paper has no named computational anchor. Proposition 2.1 is an inequality; Proposition 3.1 is an impossibility about distance axioms; Theorem 3.4 and Propositions 3.5–3.7 establish consistency and asymptotic properties. None gives a complexity classification, algorithm, approximation guarantee, or parameterized result. The polynomial-time observation about Wasserstein distance is unnumbered, and the intractability of empirical \(i\)-Kemeny scores is imported from earlier work. Wrapping Theorem 3.4 in a newly invented threshold-evaluation problem does not turn it into a computational theorem of this paper.
The proposed Continuous Positionwise Distance Evaluation\(_\infty\) is mathematically legitimate, but that is precisely its weakness as a ChoCo anchor. The paper’s distance already depends on the voter multiset only through
\[ F(E)=\frac1n\sum_{v\in V}\operatorname{freq}(v). \]
Replacing this empirical average by \(\sum_t\mu_t\operatorname{freq}(t)\) changes notation, not the computational object. Clearing denominators recovers exactly the same finite frequency matrix. There is no mass-transfer decision, winner condition, feasible population intervention, separation problem, or population-dependent combinatorial structure. The candidate matching and Wasserstein computation are unchanged. Theorem 3.4’s UN-consistency is a statement about the resulting matrices and candidate counts; it is not a phenomenon driven by voter multiplicity.
The stronger variants do not repair the provenance problem. With explicit ballot support, the continuous evaluator is a weighted scan followed by standard matching. With a succinct generator or sampling oracle, difficulty may arise in computing expected frequencies, but then it comes from the representation or access model for that generator, not from continuizing the society. A continuous latent preference space likewise pushes forward to the same ballot distribution. Retaining named voter identities would not help, because the distance is deliberately invariant to voter reordering; doing so would define a different distance.
One could instead continuize DAP and study exact weighted \(i\)-medoid computations. That might be a worthwhile new problem, but it is not a named computational result here: the paper uses heuristic computation for these features and states no complexity theorem about them. Likewise, continuizing the MDS/KK map would concern a second-level dataset of elections, not the population axis of an individual election.
Thus the strongest negative verdict is that this paper should fail ChoCo’s strict computational-anchor test. The high-multiplicity regime itself is sensible, and the proposed weighted distance is a natural Class-A extension; it would be wrong to reject it because masses are “not real voters” or because the answer is easy. But it is a new, modest evaluation problem manufactured around a structural theorem, not a continuous mirror of computational content supplied by the paper. The universal claim that no related continuous research could ever be worthwhile is therefore stronger than the evidence supports; the defensible claim is that this paper does not provide a qualifying 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.