| paper | d4-journals-jet-2005-0032 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | no |
The numbered Theorems 1–3 establish existence and structural properties of farsighted stability, but state no algorithm, complexity bound, or hardness result. The proposed continuous membership problem adds a computational question; it cannot supply the missing named computational anchor.
fails bit a — no named computational result to mirror
The proposed type-based model concerns farsighted consistency for aggregate collaboration networks; it does not mirror a computational result in the paper. Section 6 reports calculations for three-agent examples, but these are unnumbered and do not provide a qualifying anchor.
The strongest positive direction is a new problem about farsighted stability in a large collaboration network. Imagine a research platform with many agents but few complete role types—say, field and institution—with every member of a type sharing the same payoffs and move rights. Type \(r\) has mass \(\mu_r\). In an aggregate model where utilities and legal moves depend on type-to-type link densities, a network state is \(x\in X\subseteq[0,1]^{R\times R}\). A coalition move changes \(x\); farsightedness follows the paper’s rule: each coalition that moves must prefer the eventual state to the state it moved from.
This gives a concrete candidate problem, Type-Continuum Farsighted Consistency. An instance consists of rational type masses, a compact semialgebraic state space \(X\), rational utility functions \(u_r(x)\), a semialgebraic relation specifying legal moves by coalitions of each type, and a target state \(x_0\). Define \(F^*\) by the paper’s consistency condition: after any legal deviation from a state in \(F^*\), some state in \(F^*\) must be reachable from the deviation such that the original coalition does not all prefer it to the starting state. The question is whether \(x_0\in F^*\). A large collaboration platform with many researchers in a few roles is a plausible high-multiplicity regime. The solution is a membership answer about the stable set, not an optimization objective.
The closest named results are Theorem 1, proved in the paper: every supernetwork over a finite collection of directed networks is RR-inductive, meaning every farsighted domination path has a network reachable from every network on that path. Theorem 2, obtained by applying Berge’s result to Theorem 1: every such supernetwork has a farsighted basis. Theorem 3, proved in the paper: for any farsighted basis \(B\), its associated stable set \(B^*\) satisfies \(B\subseteq B^*\subseteq F^*\); the largest farsightedly consistent set \(F^*\) is nonempty and externally stable under farsighted dominance. These are existence and stability statements. None asserts that a computational problem is in P, NP-hard, FPT, or another complexity class. The two three-researcher calculations in Section 6 are unnumbered examples, not qualifying anchors.
So the candidate problem is a plausible extension of the paper’s stability questions, but it is not an anchor-based continuous mirror of a named computational result. Theorems 1 and 2 rely on the finite set of networks, and their argument does not establish existence or computability for a continuum of states. I would expect the proposed membership problem to raise continuum-specific difficulties, but the paper gives no basis for predicting its complexity. Further questions include whether \(F^*\) exists under these continuum moves and what restrictions make it computable.
The soft spot is whether type-level link densities retain enough of the paper’s network structure. Its agents have links to particular named agents, and payoffs can depend on the entire network. Aggregate densities may erase relevant topology; preserving it with a graphon-like state would make the model richer and the computational input harder to specify. That leaves a credible new research direction, but no honest qualifying anchor in this paper.
The strongest negative is that none of the three anchors is a computational result. Theorem 1 proves RR-inductivity from finiteness: along an infinite path through a finite set of networks, some network repeats. Theorem 2 applies Berge’s existence theorem, and Theorem 3 relates stable sets and establishes nonemptiness. None gives an algorithm or a complexity bound. Asking for a continuous analogue of those existence statements would be an analytic question; asking whether a state \(x_0\) belongs to \(F^*\) adds a computational problem that the paper does not state. So Theorems 1, 2 and 3 do not supply the required anchors.
The best rescue is the paper’s computation of largest consistent sets in Section 6, or its remark that computing a farsighted basis remains open. But a continuous version needs a finite, effective description of its states and moves. The proposed type-to-type link densities \(x\) may not suffice: the paper allows payoffs to depend on the entire network, and its deviations change particular arcs through coalitions. Two networks with identical link densities can have different topology and different reachable outcomes. A graphon-like state could retain more structure, but then the proposal needs to say how that object, its legal moves and the relevant paths are finitely encoded. A fixed block model can provide such an encoding only by adding restrictions on payoffs and transitions. The proposed semialgebraic model stipulates a new problem; it does not yet show that the paper’s computation has a natural continuous input model.
That is the strongest case against a worthwhile *mirror* here, but it does not establish that every continuous network-formation problem would be uninteresting. A large collaboration network with repeated roles is a plausible high-multiplicity regime; treating type masses as the issue would be a mistake. The narrower conclusion is that the paper offers no named computational result to mirror, and the computational rescue requires enough new modeling choices that it is better described as a separate research direction.
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.