| paper | d4-conf-aaai-2014-0384 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | unclear |
No numbered theorem, lemma, corollary, or proposition states a computational guarantee or complexity result; the algorithms and experiments do not meet the required named-result gate. The proposed population-average objective is statable, but it changes the paper’s single-agent objective and is unanchored, so it cannot supply the required mirror.
fails bit a — no named computational result to mirror
The paper presents utility propagation algorithms and experiments for finding optimal contracts, but no qualifying numbered computational result is available to carry over.
The paper has no qualifying anchor under your rules. It contains no numbered Theorem, Lemma, Corollary, or Proposition asserting that a problem is in P, NP-hard, FPT, or otherwise classifying its complexity. Its numbered items are algorithms: for example, “Algorithm 1: Main Algorithm” describes utility propagation. The paper reports experiments comparing that method with simulated annealing, but those are not a named complexity result. I can’t honestly restate an anchor that isn’t there.
The strongest plausible mirror I can offer is therefore unanchored, and it changes the paper’s single-agent objective into a population objective:
Continuous Population Optimal Contract. An instance consists of a finite set of issues \(I\), each with a finite domain \(D_i\); a finite support \(T\) of preference types; and binary-rational masses \(\mu_t>0\) summing to \(1\). Each type \(t\) fully specifies a constraint-based utility function \(u_t(c)\), represented by its hypergraph factors, scopes, and local utility functions over the issue domains. The question is to find a common contract
\[
c^\star\in\prod_{i\in I}D_i
\quad\text{maximizing}\quad
\sum_{t\in T}\mu_t u_t(c^\star).
\]
A solution is any maximizing contract.
A plausible regime is a large federation choosing one shared multi-issue contract, where many members fall into a small number of standardized requirement profiles. A type must include the full utility representation—issue domains, constraint scopes, and utility parameters—so members counted under the same type are genuinely indistinguishable for this decision. The population may be large while the number of supported profiles \(\tau=|T|\) remains modest; if preferences are bespoke, \(\tau\) may instead approach the population size.
The paper’s best point of contact is its stated task of finding optimal contracts for a preference profile, using a utility function decomposed over issues and constraints. The proposed question keeps the paper’s finite issue domains and hypergraph utilities, while treating the population’s average utility as the preference profile to optimize. Its scope would be the underlying common-contract search—not the message-passing encoding, negotiation dynamics, or future hierarchical protocol.
My expected direction is Class B if the corresponding finite contract-optimization problem is hard: the continuous formulation includes the one-type case \(\mu_t=1\), so any hardness of that finite problem would carry over. The paper itself does not establish that hardness. In fact, averaging the types simply produces another contract utility function, \(\bar u(c)=\sum_t\mu_tu_t(c)\); the chosen contract still comes from finite domains. So I would not expect continuization alone to remove difficulty caused by the issues or hypergraph structure. It may still be tractable on suitably restricted structures, but this paper does not prove that either.
The mirror is plausible but not secure as a mirror of the authors’ exact problem: the paper describes an agent maximizing that agent’s utility, while this proposal maximizes population-wide average utility. A shared contract and a standardized preference population make that extension reasonable, but it changes the objective. That is the weakest point, alongside the absence of any qualifying named computational result.
There is no named computational result here to anchor a mirror. The numbered items are algorithms, not theorems about complexity; the paper reports experiments on generated instances and compares runtimes, but gives no complexity classification or algorithmic bound. Under the requested standard, that leaves no qualifying anchor. The proposed population question is therefore an unanchored extension, not a continuization of a result in the paper.
A federation choosing a shared contract is plausible, and its members could be grouped by complete utility hypergraphs. So lack of multiplicity is not the objection. The problem is what the population is supposed to optimize. The paper finds contracts maximizing one agent’s utility. Maximizing population-average utility replaces that objective with a social welfare rule and assumes utilities are comparable across agents—choices the paper neither makes nor studies.
Even granting that extension, let \(u_t(c)\) be type \(t\)’s utility for contract \(c\). Then the proposed objective is
\[
\sum_{t\in T}\mu_t u_t(c).
\]
Because each \(u_t\) is a sum of local constraint utilities, this is itself a single utility function: combine the types’ factors and weight each factor by its type’s mass. The resulting task is the paper’s original single-agent contract search on a composite hypergraph. Population size has disappeared; support size and masses describe the encoding of that hypergraph. That is a clean application, but it does not extend a computational result about populations, since the paper establishes no such result and the aggregation removes the population as a distinct computational object.
A stronger alternative could preserve disagreement by asking for a contract acceptable to at least \(q\) fraction of the population, or for a Pareto frontier across types. Those are meaningful group decision problems, but their acceptance rule or multiobjective criterion is new. Returning each type’s own optimal contracts makes the masses irrelevant; assigning different contracts to different types requires coupling constraints absent from the paper. So the nontrivial variants change the central task rather than continuize it.
This negative case has a real limit: the shared-contract setting is sensible, and the average-utility problem is well-posed. I would reject it as a worthwhile *mirror of this paper’s results*, not claim that no valuable follow-on problem can be built from the model.
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.