| paper | d4-conf-aaai-2015-0751 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | medium |
| authors would recognise it | no |
Theorem 2 is a genuine computational result, but its reachability question depends on individual voters making unilateral best responses. In the nonatomic limit, an individual cannot change the plurality outcome; the proposed cohort moves restore influence by changing who acts and how, so they do not preserve the anchored problem.
fails bit b-changes — the continuum limit changes the question
The cohort update is a coordinated group action rather than an individual best response; the poll scenario makes that alternative model plausible but does not establish that it is the authors’ reachability problem.
fatal: True
The proposed mirror covers only Theorem 2’s basic-model reachability result; it leaves the truth-biased and lazy-voter algorithms and other characterisation results aside.
The strongest case I can defend is narrow: the basic plurality model’s target-profile reachability result, Theorem 2. The theorem is proved in this paper, not merely cited. It says that, given a truthful profile \(a\) and a distinct profile \(b\), deciding whether \(b\) can be reached from \(a\) by iterative best-response updates is NP-complete. The paper gives a polynomial-length path certificate and proves hardness by reduction from Hitting Set.
A plausible high-multiplicity setting is a very large membership poll over a small slate of choices, where members have only a modest number \(\tau\) of recurring preference rankings. For example, a platform might let members revise plurality votes for convention locations after seeing interim results. Here \(N\), the number of members, can be far larger than \(\tau\). A type \(t\) is a complete ranking of the choices, and its mass \(\mu_t\) is the fraction of members with that ranking.
The continuous question I would hand to a prover is Plurality Cohort Reachability. An instance has candidates \(C\) with the paper’s lexicographic tie-breaking, rational type masses \(\mu_t\), and a rational target state \(\hat{x}\). A state \(x_{t,c}\) records the mass of type \(t\) currently voting for candidate \(c\); initially, everyone votes for their top-ranked candidate. Candidate \(c\)’s score is \(S_c(x)=\sum_t x_{t,c}\).
A move chooses a type \(t\), a candidate \(c\) currently receiving some of its mass, and a rational amount \(0<\delta\le x_{t,c}\). That cohort moves its votes from \(c\) to \(d\), where \(d\) is a best response for the cohort against the remaining population \(x-\delta e_{t,c}\): it maximizes \(u_t(F(x-\delta e_{t,c}+\delta e_{t,d}))\) over candidates. The move must strictly improve the winning candidate for type \(t\), compared with \(F(x)\). The question is whether \(\hat{x}\ne x^0\) is reachable from the truthful state \(x^0\) in finitely many such moves. A solution is a finite list of type, source, destination, and mass choices.
This mirrors the paper’s reachability question itself, rather than an encoding of it. It covers Theorem 2’s basic-model result only; I would not stretch it to the truth-biased or lazy-voter algorithms, whose incentives depend more directly on individual pivotality and abstention.
My tentative expectation is Class A, though it is a conjecture about this mirror, not a consequence of Theorem 2. The theorem’s Hitting Set reduction appears to use indivisible choices of voters. Splittable mass may relax that integrality and permit a flow or LP characterization. But the order of winner changes matters, so it is an open question whether the path can actually be summarized by such a formulation. The first follow-up questions are whether finite reachability has a useful path-length bound, and whether exact discrete instances embed without fractional moves creating extra paths.
The soft spot is substantial: a positive-mass cohort updating together is a macro-player. In a genuinely nonatomic population, a single agent has zero influence on the plurality outcome, so the paper’s unilateral best-response dynamics largely disappear. Cohort updates make the dynamics nontrivial, and could fit a poll where whole preference segments revise in response to the same interim result, but the authors or a referee might regard that as a different dynamic rather than their problem’s continuous form. The case for the mirror rests on that batch-update interpretation being accepted.
The proponent has chosen the paper’s strongest anchor: Theorem 2 is a genuine computational result, and a large poll with a small number of recurring preference types is a plausible high-multiplicity setting. The difficulty is that the proposed mirror changes the theorem’s move rule at its core.
In the paper, one voter changes one ballot. Under plurality, that voter can strictly improve only when their individual vote changes the winner. In a nonatomic society, an individual has measure zero: changing their ballot does not change any score or the lexicographic winner. So the direct continuous limit has no strict best-response moves, and every target other than the truthful state is unreachable. The proposed move of an arbitrary positive mass \(\delta\) avoids that collapse by letting a cohort move together. But that is a coalition acting on its shared interest, not an individual making a best response. Having the same type makes agents interchangeable in the state description; it does not make them coordinated.
A more faithful alternative would ask about limits of reachable finite-population paths. That runs into a boundary problem: whether a voter is pivotal can depend on a score margin of one vote, or \(1/N\) after normalization. Such margins disappear in the limiting distribution. Two sequences of finite profiles can converge to the same continuous state while differing in those margins, and hence in which first moves are available. Recovering that information requires retaining finite-scale offsets; discarding it requires choosing which approximating sequences count. Either choice adds structure that the population distribution alone does not supply.
One can certainly define a cohort-dynamics problem, but its reachability relation would depend on additional choices: which masses may coordinate, how large a move must be to change the winner, and how cohorts are scheduled. Those choices are not consequences of high multiplicity. The resulting problem might be worth studying, but it would analyze a new collective-update model rather than continue the paper’s unilateral best-response problem.
That is a substantial case against this anchor, but it does not establish the universal claim that no worthwhile mirror exists. If coordinated preference segments are accepted as the relevant actors in a poll, the cohort version could be a sensible computational question. The paper gives enough to make that possibility hard to rule out; the negative case is strongest against treating Theorem 2’s reachability result as having a canonical continuous mirror.
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.