| paper | d4-conf-ecai-2024-0036 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | medium |
| authors would recognise it | no |
Theorem 5(d) is a genuine computational hardness result, and recurring weight classes give a plausible high-multiplicity regime. But the proposed question compares renormalized asymptotic coefficients after adding a vanishing fraction of players, rather than continuing the theorem’s exact index-maintenance decision over population mass. That is a worthwhile asymptotic question, but this anchor does not yield a surviving ChoCo mirror.
fails bit b-changes — the continuum limit changes the question
The entrants number only \(O(\sqrt N)\), so their share tends to zero; moreover, the proposal replaces exact finite-game equality with equality of a leading asymptotic coefficient, which can discard lattice and rounding effects.
fatal: True
The proposal addresses only the Penrose–Banzhaf maintenance branch of Theorem 5(d); it leaves the Shapley–Shubik branch and the paper’s other control results aside.
My lead is a mirror of Theorem 5(d), proved in the paper’s appendix. It states that, for either the probabilistic Penrose–Banzhaf index \(\beta\) or the Shapley–Shubik index \(\varphi\), CONTROL-BY-ADDING-PLAYERS-TO-MAINTAIN-\(\gamma\) is NPPP-complete. In the underlying problem, the instance is a weighted voting game \(G\), a pool \(M\) of players specified by their weights, a distinguished player \(p\), and a bound \(k\). The question is whether adding at least one and at most \(k\) players from \(M\) leaves \(p\)’s chosen power index exactly unchanged. The paper proves the hardness here, by reduction from E-EXACT-SAT; I focus on its \(\beta\) case.
A plausible high-multiplicity regime is a large member-based federation or cooperative whose voting rights come in a small number of standard weight classes. A type is a player’s voting weight, together with whether that player is an incumbent or eligible to join; \(p\) is one marked incumbent. The incumbent population has \(N\) players and type proportions \(\mu_t\), with \(\tau\) fixed or moderate while \(N\) grows. New members can arrive in cohorts of the eligible types. This is a particular regime for weighted voting games, not a claim that every legislature or shareholder group has high multiplicity.
Here is a precise continuous question for that regime, which I’ll call Local-limit Banzhaf Maintenance. Its input is a finite set \(T\) of incumbent types with positive integer weights \(w_t\) and rational proportions \(\mu_t\) summing to \(1\); a distinguished player of weight \(w_p\); eligible joining types \(u\) with positive integer weights \(v_u\) and available cohort bounds \(\eta_u\); a cohort budget \(\kappa\); and a rational quota offset \(\theta\). Assume the incumbent weights have greatest common divisor \(1\), and write \(\bar w=\sum_t\mu_t w_t\) and \(\nu=\sum_t\mu_t w_t^2\).
Interpret these data by taking \(N\) incumbent players with exactly \(N\mu_t\) of each type, and setting the quota to \(q_N=\lfloor N\bar w/2+\theta\sqrt N\rfloor\). It stays fixed when members are added. A continuous solution is a vector \(x=(x_u)\) with \(0\le x_u\le\eta_u\) and \(\sum_u x_u\le\kappa\): in the finite approximants, \(x_u\) represents adding about \(x_u\sqrt N\) players of type \(u\), with at least one cohort added overall. The question is whether such an \(x\) keeps the leading-order Banzhaf power of \(p\) unchanged.
The finite Banzhaf index tends to zero in this scaling, so the continuous question uses its leading-order value \(\mathcal B(x)=\lim_{N\to\infty}\sqrt N\,\beta_N(x)\). The local limit theorem gives \(\mathcal B(x)=\frac{2w_p}{\sqrt{2\pi\nu}}\exp\!\left(-\frac{2(\theta-\frac12\sum_u x_uv_u)^2}{\nu}\right)\). Thus a solution must satisfy \(\mathcal B(x)=\mathcal B(0)\). With positive joining weights and a nonzero cohort, this is equivalent to \(\sum_u x_uv_u=4\theta\). The expected direction is tractable: feasibility is a linear program. In fact, for \(\theta>0\), it suffices to check whether \(4\theta\) is no greater than the maximum of \(\sum_u x_uv_u\) over the budget and supply bounds; for \(\theta\le0\), no positive addition can preserve this leading-order value.
This is a continuous population model: \(\mu\) describes the society, and the decision is a real mass of potential entrants by type. The \(\sqrt N\) scale is a co-modification forced by the index: Banzhaf pivotality depends on coalitions close to the quota, whose random weight fluctuates on that scale. It keeps the original game, quota, distinguished player, addition operation, and probabilistic power measure in the finite approximants, while making the type-level admission decision continuous.
The main weakness is that this mirrors only a narrow, near-balanced asymptotic regime of Theorem 5(d). It rescales the index and imposes a specific quota and cohort scale; the paper’s reduction instead relies on carefully encoded weights and exact coalition counts. A referee might regard this as a useful asymptotic continuation of their problem rather than a mirror of the theorem’s unrestricted NPPP-complete problem. Still, the theorem’s hardness does not rule out this regime: a large society with a few recurring weight classes is a sensible high-multiplicity object, and the paper’s result does not classify its local-limit maintenance question. This case covers the \(\beta\)-maintenance branch of Theorem 5(d), not the \(\varphi\) branch or the paper’s other control results. It also suggests follow-up questions about increase, decrease, and nondecrease in the same limit, and about what happens outside the near-quota regime where the local Gaussian approximation applies.
The NPPP result in Theorem 5(d) is a real computational anchor, and repeated voting weights are a plausible high-multiplicity regime. So I would not argue that weighted voting has no sensible continuous population model. My objection is narrower: the proposed Local-limit Banzhaf Maintenance does not continue the population-level control decision in that theorem.
The entrant coordinate \(x_u\) represents about \(x_u\sqrt N\) players. As a fraction of the enlarged society, that is \(x_u/\sqrt N\), which tends to zero. The post-entry type distribution therefore converges to the incumbent distribution \(\mu\) for every feasible \(x\). What survives is a second-order perturbation to a Gaussian approximation of one distinguished player’s pivot probability. Correspondingly, the maintenance condition reduces to \(\sum_u x_uv_u=4\theta\): the type mixture matters only through a variance parameter, and the decision is a scalar weighted-sum feasibility check. That can be a legitimate asymptotic power-index question, but it is not the mass-addition control problem suggested by a continuous society.
Scaling entrants proportionally to \(N\) would make their mass visible in the limit, but with the quota fixed as in the paper, their positive total weight shifts the random coalition’s mean by order \(N\), while its standard deviation is order \(\sqrt N\). In the proposed near-majority regime, the pivotal event then becomes exponentially rare. The ordinary limiting index is zero before and after, so maintenance is vacuous. Rescaling to compare large-deviation rates might produce a separate question; moving the quota with the entrants would avoid the degeneration but would change the paper’s control operation. Neither recovers this theorem’s population-mass maintenance problem.
There is also a real semantic gap: the theorem asks for exact equality of finite-game indices after adding players. The proposal asks whether two leading asymptotic coefficients agree. That is a reasonable relaxation, but finite equality can depend on lattice and rounding effects that the coefficient discards. Calling the relaxation a mirror therefore needs more than the fact that it preserves the same index name and addition story.
This is enough for me to oppose this particular construction as a ChoCo mirror, while conceding its value as a focused asymptotic analysis of Banzhaf power. It is not enough to sustain the universal claim that no worthwhile mirror exists in any scenario: the paper has genuine complexity results, and recurring weight classes are credible. The negative case is strongest on the mismatch between vanishing entrant mass and the programme’s population continuization; beyond that, it is weak.
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.