| paper | Replicating Electoral Success |
| authors | Kiran Tomlinson, Tanvi Namjoshi, Johan Ugander, Jon Kleinberg |
| venue | AAAI 2025 |
| filed under | voting · theory |
| judged by | gpt-5.6-luna / xhigh (triple__luna__xhigh__c2r1) |
| judge confidence | high |
| authors would recognise it | yes |
The paper’s numbered results are analytic statements about dynamics and equilibria, not named computational results. The finite-type prediction problem is a recognizable computational extension of the model, but it is newly introduced and cannot supply the required computational bit. Therefore the paper is red despite offering a potentially worthwhile follow-up project.
fails bit a — no named computational result to mirror
The proposed finite-type prediction problem is a new computational wrapper around the dynamics: the paper never states it, and \(\mu\) is only a fixed vote-share input rather than part of a classified computational problem.
fatal: True
The proposed extension covers the plurality winner-copying dynamics and the \(k \le 4\) versus \(k \ge 5\) convergence dichotomy; it leaves simulation variants, atom-based equilibrium results, general asymptotic classification, and computational complexity untouched.
The strongest honest conclusion is that this paper does not contain a qualifying computational anchor under the ChoCo rule. The numbered results are analytic statements about replicator dynamics, convergence, non-convergence, and equilibria. Theorem 2 gives an exact formula, Theorem 5 proves non-convergence, and Theorem 14 characterizes some equilibria, but none asserts that a named problem is in \(P\), NP-hard, FPT, or otherwise computationally classified. All are proved in this paper; none is a cited complexity result.
So there is no compliant anchor-specific mirror to defend. The paper is already continuous in several senses: it has a continuum of voters, continuous policy positions, and distributions over candidate positions. But its contribution is analytic, not computational. Treating Theorem 2 as a \(P\)-result would be an inference made by us, not a computational result stated by the authors.
The closest positive construction would be the following relaxed-anchor extension of Theorem 2.
Call it Finite-Type Replicator Prediction\(_\infty\). There is a finite policy grid \(P=\{p_1,\ldots,p_r\}\), a finite set \(T\) of ideological voter types, and a rational population vector \(\mu\in\Delta(T)\). A type specifies a complete preference over candidate positions in \(P\), including its tie convention. Thus \(\mu_t\) is the fraction of a large repeated electorate having type \(t\). There is also a rational distribution \(\nu_0\in\Delta(P)\) of candidate-position types in the initial generation.
For a \(k\)-candidate profile \(\mathbf p=(p_{i_1},\ldots,p_{i_k})\), plurality vote shares are computed from \(\mu\). Ties among maximum-share candidates are broken uniformly. Define
\[ R_{\mu,k}(\nu)(p) = \sum_{\mathbf p\in P^k} \left(\prod_{j=1}^k \nu(p_j)\right) \Pr\!\left[ \operatorname{Plur}_{\mu}(\mathbf p) \text{ has position }p \right]. \]
The instance consists of \(P,T,\mu,\nu_0,k\), a horizon \(H\), and a centrality tolerance \(\delta\). The task is to output the exact distribution
\[ \nu_H=R_{\mu,k}^{\,H}(\nu_0) \]
and decide whether
\[ \nu_H\!\left([1/2-\delta,1/2+\delta]\right)\ge 1-\varepsilon. \]
The decision variable is therefore the predicted candidate-position distribution, and the objective is to determine whether winner-copying drives the population toward the centre or toward ideological separation.
The intended regime is many repeated local electorates and many repeated candidate slots, with \(N\gg |T|\) voters and \(M\gg |P|\) candidate observations, while the number of ideological and policy types remains modest. Rational masses are exactly clone multiplicities after denominator clearing. The scenario is recognisable to the authors: one-dimensional Euclidean voters, plurality, \(k\) candidates, and candidates copying successful previous positions.
For fixed \(k\) and finite horizon, this version is plausibly Class A: direct enumeration of the \(r^k\) candidate profiles gives an exact update, and Theorem 2 suggests a much more compressed treatment for \(k=2\). The asymptotic question—whether the iterates converge to the centre or retain mass away from it—is less clear. Theorems 1 and 5 suggest a sharp \(k\le4\) versus \(k\ge5\) boundary, but they do not establish the complexity of deciding that property for an encoded finite-type instance. That could become a genuinely continuum-specific or representation-specific problem.
The weakest point is decisive: finite policy support introduces atoms and positional ties, whereas the paper’s theorems assume atomless candidate distributions; and the paper’s main difficulty lies in the candidate-position dynamics under a fixed uniform voter continuum, not in voter multiplicity. This proposal is therefore an author-recognisable extension, but not a direct high-multiplicity mirror of a named computational result. I would present it as a promising computational follow-up inspired by Theorems 2 and 5—not as a successful ChoCo anchor.
The negative case is unusually strong at the programme’s actual boundary: this paper contains no qualifying computational anchor. Its numbered results are analytic statements about iterated distributions, convergence, non-convergence, and equilibria. Theorem 2’s formula, Theorem 5’s non-convergence theorem, and Theorem 14’s equilibrium list specify no encoded computational problem, decision version, complexity bound, or algorithmic classification. Thus there is no computational result here whose high-multiplicity relaxation ChoCo could study.
The proposed “Finite-Type Replicator Prediction” does not repair that gap. It is a new computational problem inspired by Theorem 2, not a mirror of Theorem 2. The theorem gives a closed-form identity for a particular analytic dynamical system; it does not assert that computing a finite iterate, or deciding a centrality threshold, is a named computational task. The distinction matters because ChoCo is meant to chart the complexity consequences of continuizing an existing computational problem, not to retrofit a complexity problem onto any continuous theorem.
The construction also assigns the wrong role to multiplicity. In the paper, the voter continuum is a fixed measure used to calculate vote shares, while the evolving object is the probability law of candidate positions across repeated elections. Replacing the voter measure by \(\mu\) does not create a new high-multiplicity computational regime: once \(\mu\) is given, the population size itself disappears from the update map. The proposed large \(N\) is therefore only a law-of-large-numbers interpretation, not the source of the problem’s computational content.
A more faithful version would retain continuous policy positions and represent \(\nu\) by a density or another finitely encoded measure, then ask for \(R_{\mu,k}(\nu)\), its iterates, or its asymptotic behavior. But that is simply an algorithmic treatment of the replicator dynamics already defined by the paper. It continuizes candidate strategies or repeated-election outcomes, not a society of voter types in the ChoCo sense. Conversely, putting positions on a finite grid makes exact computation possible but turns the problem into a newly designed finite dynamical system with atoms and positional ties. A recognized tie convention can fix the formalism; it cannot make the resulting problem an authored computational result or a genuine high-multiplicity relaxation of one.
The same objection defeats attempts to use Theorems 5, 10, or 14 as stronger anchors. Deciding whether an encoded initial distribution converges to the centre, retains a central gap, or lies in a basin of attraction could be worthwhile dynamics research. But it would be a new stability or prediction problem. Nothing in the paper establishes its computational status, and no choice of type space changes that fact.
So the strongest defensible negative verdict is: this is a continuous analytic paper, not a computational social-choice paper with a continuous mirror. I would not claim that every computational follow-up inspired by it is scientifically worthless; the proposed dynamics problem could be a legitimate separate project. But it does not qualify for the ChoCo programme, and the universal claim is strongest as a programme-scope judgment, not as a theorem that no useful computational reformulation could ever be invented.
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.