| paper | d4-conf-atal-2014-0102 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | medium |
| authors would recognise it | yes |
other→P discrete result → its continuous mirror
Answer: Polynomial-time computable. · discrete side: other · adjudication: proved_with_minor_gaps
full proof & adversarial review →
Theorem 4.1
statement extracted from the paper’s text layer
Given a finite set of decisions \(D\), goals \(G\), and a rational-mass distribution \(\mu\) over supported types \(t\), where each type is a complete pair of decision frameworks with trust scores, compute all decisions \(d\) for which no \(d' \ne d\) has \(q_\mu(d',g) \ge q_\mu(d,g)\) for every \(g \in G\), with strict inequality for at least one goal, where \(q_\mu(d,g)=\sum_t \mu_t v_{t,d,g}\) and \(v_{t,d,g}=1\) exactly when \(d\) meets \(g\) in type \(t\)'s trust-resolved joint framework.
A society of repeated two-agent panels, represented by a distribution \(\mu\) over complete framework-and-trust-score types. Each type contributes binary goal coverage \(v_{t,d,g}\); mass-weighted coverage \(q_\mu(d,g)\) gives each decision a population goal-coverage vector, and the output is the set of Pareto-maximal decisions.
The paper compares goal sets within one joint framework, while the proposed rule compares average goal coverage across different frameworks; the single-type boundary does not uniquely justify that mixed-population criterion.
fatal: False
Covers the dominant-decision computation characterized by Theorem 4.1; it does not continuize the dialogue protocol, its termination and success results, the trust-free conflict variants, or the software evaluation.
There is a narrow positive case, but the paper does not contain a theorem classifying a problem as polynomial-time solvable or hard. My anchor is Theorem 4.1, proved in the paper: for the two agents’ trust-resolved joint decision framework, a decision \(d\) is jointly dominant exactly when the argument \(\{sel(d)\}\vdash sel(d)\) belongs to an admissible set in the union of their dominant ABA frameworks. The problem being computed is which decisions are jointly dominant; the ABA argument is the paper’s characterization of that computation.
A plausible high-multiplicity setting is a network of many regional clinical panels, each made up of two cooperating hospitals. They all consider the same finite set \(D\) of treatments or medical references, using a shared vocabulary of attributes \(A\) and goals \(G\). A panel’s type is its complete pair of decision frameworks, including their information tables and trust scores. Panels of the same type are indistinguishable in every respect this model uses. The society is a distribution \(\mu\) over the \(\tau\) supported types, where \(\mu_t\) is the fraction of panels of type \(t\). For example, thousands of panels might use a few dozen recurring assessment templates.
Here is the continuous problem I would hand to a prover: Population-Dominant Decision. For each type \(t\), form the joint framework \(F^J_t\) using the paper’s trust-based merge. Let \(v_{t,d,g}=1\) if \(d\) meets goal \(g\) in \(F^J_t\), and \(0\) otherwise. Define the fraction of panels whose framework says \(d\) meets \(g\) as
\[
q_\mu(d,g)=\sum_t \mu_t v_{t,d,g}.
\]
Return every decision \(d\in D\) for which there is no \(d'\ne d\) satisfying \(q_\mu(d',g)\ge q_\mu(d,g)\) for every \(g\), with strict inequality for at least one goal. The decision variable is the chosen \(d\); the criterion is Pareto-maximal coverage across the population, rather than a scalar objective.
This is an exact extension at the single-type boundary: if \(\mu\) puts all mass on one panel type, each coverage is either \(0\) or \(1\), and the returned decisions are precisely that type’s jointly dominant decisions under the paper’s definition. Thus the continuous question preserves the anchor’s decision criterion while allowing fractional, population-level goal coverage. It continuizes the population of decision-making units; the candidate outcomes remain finite.
I expect this mirror to be tractable: with the type tables explicitly listed, compute the coverages and compare every pair of decisions, in time polynomial in \(\tau\), \(|D|\), \(|A|\), \(|G|\), and the rational input bit length. That is a proposed complexity result, not one Theorem 4.1 proves. A further question is whether a target decision can be made population-dominant at minimum cost, given costs for transferring mass between complete framework types.
The weak point is that the paper studies one two-agent interaction, while this mirror aggregates the coverage of many such interactions into a shared recommendation. That is a real co-modification, and the authors might prefer to study the panels separately. I would keep the claim narrow: it mirrors the paper’s dominant-decision computation, not its dialogue protocol or its full contribution. The resulting problem also looks easy, so this is a modest tractability mirror rather than an example of discrete hardness dissolving.
The case against is strongest on Theorem 4.1, but it is not airtight. The theorem is a named computational characterization: it connects jointly dominant decisions to admissible arguments. It does not establish a complexity bound or a new algorithm for that problem. Given the paper’s explicit tables, one can directly compute each decision’s goal set and compare those sets for dominance. The implementation timings in Section 6 do not add a complexity result. So the paper offers little of the algorithmic landscape ChoCo is meant to chart.
The proposed society of clinical panels is conceivable: a large health system could have many panels built from recurring templates. But a type would have to include the full pair of agents’ frameworks, including their knowledge and trust scores. If the panels share one type, the population adds nothing to Theorem 4.1. If they have different types, the theorem applies separately to each panel type.
The proposed coverage values \(q_\mu(d,g)\) create a new problem. They average goal satisfaction across panels and then apply Pareto dominance to those averages. That is not the paper’s dominance condition, which compares goal sets within one joint framework. The single-type case recovers the paper’s rule, but mixtures can combine goal coverage across different panels. That may be a sensible population-level recommendation rule; it needs its own justification, and it is not supplied by Theorem 4.1. Its computation is also immediate from the explicit support: calculate each type’s table, sum the rational masses, and compare the resulting vectors. There is no evident hard pricing problem or other obstacle for continuous-optimization methods to resolve.
The most faithful population version would preserve each panel’s local decision problem and report its dominant decisions by type, perhaps also their prevalence. That is well-defined, but it just runs the paper’s computation once per supported type and summarizes the answers. A question about the minimum mass of panels whose frameworks must change to make a decision dominant would be a further departure: transferring mass between types changes which panels, with which knowledge and trust relations, are represented. The paper gives no intervention semantics for that operation.
The same limitation is sharper for Theorem 5.1. Its object is a finite dialogue between two agents, with utterances, targets and turns. A distribution over repeated two-agent dialogues gives separate dialogues and their outcomes; it does not yield a dialogue among a continuum of agents. Extending the protocol to many agents could be worthwhile, but the paper identifies that as future work, and a continuous model would first need a new account of how those interactions work.
So I would not claim that no plausible high-multiplicity scenario exists. The negative case is weaker than that universal claim: repeated panels could support an aggregate recommendation question. But the proposed aggregate criterion is a new social-choice layer, while the faithful version is a straightforward batch of the paper’s existing computation. Neither gives a compelling continuization of the paper’s central result for ChoCo’s complexity programme.
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.