| paper | d4-journals-geb-2014-0039 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | no |
Theorem 1 and Theorem 2 are axiomatic characterizations, not computational results. The proposed limit of replicated Shapley shares would be a new computational question, not a continuous analogue of a computational result in this paper.
fails bit a — no named computational result to mirror
Theorem 1 characterizes a family of cost-sharing rules, and Theorem 2 characterizes three rules by fairness and incentive properties; neither is a computational result. The shortest-path observation is unnumbered background.
I can’t give a compliant anchor-based positive case: the paper has no numbered result asserting that a computational problem is in \(P\), NP-hard, FPT, or otherwise classifying its complexity. Theorem 1, proved here, characterizes the unique weighted Shapley cost-sharing solution satisfying Group Independence, Equal Treatment for Equal Demanders, and Consistent Treatment for Unequal Demanders, for \(\alpha\in(0,1)\). Theorem 2, also proved here, characterizes \(y^0\), \(y^K\), and \(y^1\) through fairness and incentive properties. Both are axiomatic results, not computational ones. The discussion of minimum-network costs being generally hard and the one-demander shortest-path case is unnumbered background; the latter relies on Dijkstra’s result. Neither supplies the requested anchor.
The most plausible continuous direction would be cost sharing for a large population of municipalities or districts that share a small number \(\tau\) of complete network profiles. A type would specify demand status and its edge costs to the source and other types; \(\mu_t\) would be the fraction of the population with that profile. The network decision is a minimum-cost tree connecting every supported demander type, with supported indifferent types available as relays. A precise computational question could ask, given the rational type masses and edge costs, for each type’s aggregate cost share in the limit of the paper’s weighted-Shapley allocation as finite populations replicate those types in proportions \(\mu\). That would test whether the limit exists, is independent of the replication sequence, and can be computed in time polynomial in \(\tau\) and the input bit length.
This is a conceivable population-continuous problem, but it is a new computational question built from the paper’s allocation rule, not a continuous mirror of a named computational result in the paper. Its weakest point is that the network model is organized around individually identified locations and pairwise edge costs. Collapsing people into types can change which sites are available and therefore change the network game. The paper gives no computational anchor that resolves that issue or establishes that its authors would treat the proposed limit as their problem.
The clearest objection is that neither named theorem is a computational result. Theorem 1 characterizes a cost-sharing rule by axioms; Theorem 2 characterizes three rules by fairness and incentive properties. Neither states a complexity bound or gives an algorithm for a computational problem. The paper’s remark that the network cost is “usually” hard to compute is background, not a result it establishes. Its one-demander shortest-path observation is also background. So there is no computational theorem here whose population axis can simply be continued.
The proposed rescue is to replicate complete network types and ask for the limiting aggregate Shapley shares. That is a coherent new problem, but its most faithful interpretation exposes the central mismatch: agents in this paper own locations. A location can be used as a relay because it exists, not in proportion to how much population it represents. Any positive mass of a type would therefore supply the full location, while zero mass would remove it. The network game changes discontinuously as a type’s mass approaches zero. Making location access scale smoothly with mass would require changing the model’s treatment of sites and edges, not just aggregating its existing players.
Nor does a large population at a municipality create many interchangeable network players. Each municipality contributes one particular position and set of pairwise connection costs; inhabitants do not each contribute another relay location. Under the paper’s own notion of type, which includes all costs and relevant characteristics, distinct physical locations will generally be distinct types. Grouping nearby or similar municipalities would be an approximation that needs its own model. Replicating a type exactly instead creates multiple players with the same network profile, whose effect depends on how their mutual edges and ownership of locations are defined.
One could still study algorithms for the replicated Shapley allocation, or for a high-multiplicity network optimization problem. But those would be new computational questions attached to the paper’s model, not mirrors of Theorems 1 or 2. The negative case is weakest here: a carefully specified replicated network model might yield a worthwhile algorithmic problem. What the paper offers no basis for is treating that extension as its continuous computational result; and the physical-location interpretation makes a population continuum a poor fit for its central object.
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.