d4-conf-aaai-2026-0040

This supports a narrow but credible mirror: metric-distortion voting with preference intensities over a continuous electorate. The lead anchor is Theorem 2, whose constructive rule can be implemented using mass capacities indexed by reported ballot types. Theorem 1 supplies a lower-bound boundary for the same design problem.

Let \(A\) contain \(m\) alternatives and fix \(\alpha\). A reported type is a complete ballot \(t=(\pi,\eta)\): a ranking \(\pi\) and an intensity mark \(\eta\in\{\succ,\succ\succ\}^{m-1}\). A continuous society is a unit-mass population \(X\) with a ballot map \(b:X\to T\); its input is the rational distribution \(\mu_t=\lambda(\{x:b(x)=t\})\), with support size \(\tau\). For a compatible metric \(d\), social cost becomes
\[ \operatorname{sc}_{\lambda}(a)=\int_X d(x,a)\,d\lambda(x), \]
and distortion is the worst-case ratio of this cost to the minimum cost of any alternative, over metrics consistent with the reports.

This keeps the paper’s information structure: the rule sees ballots, while the metric is hidden and adversarial. For a high-multiplicity scenario, imagine a large public consultation on facility sites, with many residents in a small number of residential or campus clusters. Residents in a cluster share the same location, costs, and reported intensity ballot; the rule uses the ballot-mass distribution, while its guarantee covers every compatible metric. More generally, residents with the same ballot may have different hidden metric positions, as they may in the paper’s model. Those latent differences must not be mistaken for identical full types; the continuous electorate can retain them while the rule sees only the ballot distribution.

Theorem 1 — proved in this paper — anchors a lower-bound question. It states that, for mandatory intensity elicitation and every \(\alpha\in[0,1]\), every voting rule has metric distortion at least
\[ 1+2\max\!\left(\alpha,\frac{1-\alpha^{\lfloor m/2\rfloor}}{1+\alpha^{\lfloor m/2\rfloor}}\right). \]
The paper proves this using two-agent instances. The continuous question is to lower-bound the optimal worst-case distortion over deterministic rules \(F\) that map rational ballot-mass distributions to alternatives:
\[ \inf_F\ \sup_{\mu,d}\ \frac{\int_X d(x,F(\mu))\,d\lambda(x)} {\min_{a\in A}\int_X d(x,a)\,d\lambda(x)}. \]
Theorem 1’s bound transfers: replace each witness voter by a block of equal mass at the same metric location and the distortion ratio is unchanged. The certificate is therefore supported on just two reported types. This is a transferred impossibility bound, not a claim of computational hardness; the exact continuous minimax value remains a further question.

Theorem 2 — proved in this paper — is the lead anchor. For \(k\in[m-1]\), it gives a rule that, on every moderate-up-to-\(k\) profile, selects an alternative with distortion at most \(2+\max(\alpha,t_k)\). Here moderate-up-to-\(k\) means no intensity appears before position \(k+1\), and the theorem’s bound uses
\[ w_1=\frac{\alpha+1}{3\alpha+1},\qquad t_1=\frac{1-\alpha}{3\alpha+1}, \]
and, for \(k>1\),
\[ w_k=1-\frac{2\alpha}{(1-\alpha)t_{k-1}+2w_{k-1}+2\alpha}, \qquad t_k=w_k+(1-w_k)t_{k-1}. \]
The continuous problem is: given \(m,\alpha,k\) and a rational mass distribution \(\mu\) supported on moderate-up-to-\(k\) ballots, find a winner with that distortion guarantee.

The paper’s PSM construction gives a concrete solution. Let \(r^k\) be its scoring vector, padded with zeros after position \(k+1\), and define
\[ q(c)=\sum_t\mu_t\,r^k_{\operatorname{rank}_t(c)}. \]
For each candidate \(a\), build a bipartite graph from ballot types \(t\), with mass \(\mu_t\), to alternatives \(c\), with mass \(q(c)\). Include an edge when type \(t\) ranks \(a\) above \(c\). Find an \(a\) and nonnegative flow \(x_{t,c}\) satisfying
\[ \sum_{c:\,a\succ_t c}x_{t,c}=\mu_t,\qquad \sum_{t:\,a\succ_t c}x_{t,c}=q(c). \]
That candidate is the PSM winner; the flow certifies the matching condition. The proof’s inequalities hold voter by voter, so they extend to integrals over a population, including hidden distance variation within a reported ballot type. With rational \(\alpha\) and masses, the flow instance has polynomial size in \(m,\tau\), and the input bit length. This is an expected Class A result for the continuous winner-selection task. The paper’s matching lemma (Theorem 3) is cited from Gkatzelis et al.; Theorem 2 and its PSM guarantee are proved here.

The authors should recognize this as their problem: it preserves single-winner voting, the same intensity semantics, and the same worst-case metric distortion objective. Only the electorate representation changes from a list of voters to mass by type. Theorem 2 raises natural follow-ups: can one compute the instance-optimal robust winner rather than a winner with a global bound? How does the algorithm extend to varying intensity ranks or a small mass of high-rank outliers, as in the paper’s later results?

The scope here is mandatory intensity elicitation, with Theorem 1’s general lower bound and Theorem 2’s moderate-up-to-\(k\) upper bound. I would not extend this case to the paper’s line-metric or price-of-ignoring-intensities results.

The weakest point is that this paper gives distortion guarantees and a constructive rule, not a named NP-hardness, polynomial-time, or parameterized-complexity classification. The polynomial-size flow formulation is a consequence of the PSM construction, not a complexity theorem claimed by the authors. Also, when masses are rational, the society can be expanded into a finite election using a common denominator. The potential contribution is thus the high-multiplicity representation and its dependence on \(\tau\), not a new voting model or a new distortion bound.