| paper | d4-journals-geb-2021-0021 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | yes |
Theorem 6 does not state a computational result, so the paper supplies no eligible result to continuize. The proposed mass-based evaluation is statable, but for the specified explicit polynomial input it is direct evaluation, and harder variants depend on an encoding the paper does not provide.
fails bit a — no named computational result to mirror
The proposed evaluation concerns only the Banzhaf reduction; it does not computationally cover the paper’s other characterizations, duality results, or induced-value results.
There is a plausible continuous population object here, but the paper supplies no eligible computational anchor. Theorem 6, proved in this paper, is the closest fit: it says that a value-free reduction is the Banzhaf reduction exactly when it satisfies additivity, null player out, permanent null player, path independence, and maximum ignorance; it also says that this reduction induces the Banzhaf value. Its formula averages \(v(S\cup T)-v(T)\) over subsets \(T\) of the removed players. That is an axiomatic characterization, not a claim that a computational problem is in P, NP-hard, or otherwise classified. The other numbered theorems likewise characterize reductions or values. So none can honestly serve as one of the required computational anchors.
The best candidate mirror I see is a large cooperative enterprise or consortium with many participants but few payoff-relevant types: participants of a type share the same technology, entitlements, and other game-relevant attributes. Let \(\mu_t\) be the mass of type \(t\), and let \(v(x)\) give the worth of a coalition containing mass vector \(x\). If the retained population is \(\nu\leq\mu\), the removed mass is \(\rho=\mu-\nu\). Under atomless independent half-probability participation, the removed coalition has mass \(\rho/2\) almost surely. The continuum counterpart of the Banzhaf reduction is therefore
\[ \Psi^{\mathrm{Ban}}_{\mu,\nu}(v)(x) = v(x+\rho/2)-v(\rho/2), \qquad 0\leq x\leq\nu. \]
A concrete follow-on computational question would be: Continuous Banzhaf Reduction Evaluation. Given rational mass vectors \(\mu,\nu,x\), with \(x\leq\nu\leq\mu\), and a rational polynomial worth function \(v\) supplied explicitly, compute the exact rational value \(\Psi^{\mathrm{Ban}}_{\mu,\nu}(v)(x)\). The solution is that value. For explicitly represented bounded-degree polynomials, the formula makes evaluation tractable by substitution. How the problem behaves for more succinct representations of \(v\) is a separate question; the paper does not settle it.
This is a recognizable continuation of the paper’s Banzhaf reduction, and the regime has a credible high-multiplicity story. But it is not a continuous mirror anchored in a named computational result of this paper, nor does the paper show that computing its reduction is hard or tractable. My weakest point is exactly that gap: a natural population-continuous version of the object exists, but turning its axiomatic study into complexity theory would add a computational programme the authors did not pursue.
Theorem 6 is the only plausible anchor the proponent raises, and it does not survive as a computational anchor. It characterizes the Banzhaf value-free reduction by axioms and shows which value the reduction induces. Like the paper’s other theorems, it says nothing about an algorithm, complexity, or even a computational problem with a specified input. There is no complexity result here for a population continuum to mirror.
A large consortium with many participants in a few payoff-relevant types is a credible high-multiplicity setting; the objection is not that such populations cannot exist. For a worth function \(v\) depending only on type masses, the Banzhaf average over coalitions of removed participants concentrates, as those populations grow, at the half-mass point. This gives the proposed expression \(v(x+\rho/2)-v(\rho/2)\). But with an explicitly represented bounded-degree polynomial, computing it is just two evaluations and a subtraction. With an arbitrary worth function, there is no finite input to compute from; with a succinct encoding, the difficulty depends on that encoding. The paper specifies none of these, and its characterization does not supply a harder task.
One could instead study the complexity of computing induced values or optimizing over retained mass vectors. Those might be worthwhile questions for a new computational cooperative-game paper. But neither is a computational result of this paper, which defines no such objective or input representation. Keeping finite-population fluctuations to retain the subset combinatorics would return to a discrete high-multiplicity problem, rather than rescue the proposed continuum evaluation.
So the negative case is strong against a *mirror of this paper*: its results are axiomatic characterizations, and the offered continuous computation collapses to direct evaluation or depends on choices the paper never makes. It cannot establish that no future model inspired by the paper would be valuable; that broader claim is inherently harder.
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.