| paper | d4-conf-wine-2023-0011 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | unclear |
No named result asserts an algorithm, computational bound, hardness result, or solver formulation, so bit (a) fails. The proposed continuous group-stability margin is explicit and related to the axiom, but it optimizes adversarial mass changes rather than the paper’s probability of axiom satisfaction after independent finite-voter noise; the proponent also labels it new work.
fails bit a — no named computational result to mirror
No computational result is covered: Theorem 5, Theorems 8 and 11, and Propositions 17–20 bound axiom satisfaction under profile noise, while the paper does not study a continuous population.
The paper contains no named computational complexity result of the kind your anchor rule requires. Theorem 5 concerns which axioms remain violated under noise; Theorem 8 gives smoothed-satisfaction and smoothed-violation rates for resolvability; Theorem 11 gives a smoothed rate for group stability; Propositions 17–20 give Mallows-noise bounds. None asserts that an underlying problem is in \(P\), NP-hard, FPT, or similar. So there are no eligible anchors, and I can’t honestly supply a continuous problem *per anchor*.
The strongest foothold for a positive case is Theorem 11, proved in this paper: for every hyperplane rule, \(o(\sqrt n)\)-group-stability is smoothed-satisfied at rate \(O_m(1/(\sqrt n\,\minprob(\mathcal{S}_\phi)^{3/2}))+o(1)\). Its subject has a natural population interpretation: how much of a society must change its reported ranking before the outcome can change?
A plausible new computational question would be Continuous Group-Stability Margin. Take a large electorate with a modest number of distinct rankings: let \(T\) be its supported ranking types, with rational masses \(\mu_t\) summing to \(1\), and let \(R\) be a positional scoring rule. Assume \(R(\mu)\) has a unique winner. A decision variable \(x_{t,t'}\ge 0\) moves mass from ranking \(t\) to ranking \(t'\), with \(\sum_{t'}x_{t,t'}\le\mu_t\). The resulting society is
\[
\mu'_u=\mu_u-\sum_{v}x_{u,v}+\sum_t x_{t,u}.
\]
The task is to find the infimum of \(\sum_{t,t'}x_{t,t'}\) subject to \(R(\mu')\ne R(\mu)\): the smallest population fraction whose changed rankings can alter the winner set. Types are complete rankings, mass is electorate share, and the action and objective are explicit. In a regime with \(n\gg |T|\), this is a sensible high-multiplicity scenario, and the question closely follows the paper’s group-stability notion. For positional scoring rules, the winner conditions are linear; optimizing over destination rankings may admit a pricing formulation, making tractability a reasonable conjecture.
That is a credible research extension, but it is not a faithful mirror of Theorem 11. The theorem studies probabilities over finite profiles after independent random perturbations. In a continuum population, those perturbations produce a deterministic aggregate distribution \(\mu K\); the finite-sample fluctuations behind the theorem’s rate disappear. Also, the theorem’s coalition range \(o(\sqrt n)\) corresponds to a population share tending to zero, so it does not itself establish a result about a fixed positive mass threshold.
Thus the best positive case is that this paper offers a recognizable axiom and a plausible population-scale optimization problem to study. Its results do not provide the required computational anchor, and the proposed margin problem should be presented as new work rather than as a continuous mirror of a named complexity result.
The strongest case against a mirror is that this paper proves no named computational result. Theorem 5, Theorem 8, Theorem 11, and Propositions 17–20 concern whether axioms hold after random perturbations and at what rates. They do not analyze the complexity of computing an intervention, winner, or probability. The proponent is right that Theorem 11 is the closest foothold, but it is an analytic statement about a finite random process, not a computational problem awaiting continuization.
The gap matters for the proposed Continuous Group-Stability Margin. At finite \(n\), independent ranking noise makes the post-noise histogram random, and Theorem 11 uses its fluctuations near decision hyperplanes. For a continuum society \(\mu\), the same noise kernel produces the deterministic aggregate \(\mu K\). Meanwhile, the theorem’s coalition size \(o(\sqrt n)\) becomes a mass share \(o(1/\sqrt n)\): away from a decision boundary, such changes eventually cannot affect the outcome; on the boundary, exact ties can make the winner set unstable. Neither case preserves the theorem’s probabilistic rate. Keeping finite-sample draws to retain that rate keeps the very fluctuation mechanism that the continuous limit removes.
The proposed margin is the strongest alternative, and it is sensible as a population problem. But it asks for the cheapest adversarial mass change, not the probability that an axiom survives noise. Its unit-cost mass transfers are a special case of the programme’s existing continuous bribery and winner-robustness model. The theorem motivates group stability, but supplies no new computational task or tool for that optimization. Extending the question to more rules or richer types may be worthwhile; that would be a new robustness project prompted by the topic, not a computational mirror of this paper’s results.
So I would reject Theorem 11 as a surviving mirror anchor, and the other named results do not supply computational anchors either. The negative case is strong for “no worthwhile continuous mirror *of this paper’s contributions*,” but weak as a universal claim: the group-stability margin remains a plausible population-scale research question if “mirror” includes a new optimization problem derived from an axiom.
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.