| paper | d4-journals-jet-2025-0080 |
| 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 is an axiomatic characterization, and the paper states no named algorithm, complexity bound, or hardness result. The proposed mixture over utility types changes the paper’s stochastic-choice rationalizability question into a new heterogeneity model, so it cannot supply the missing computational anchor.
fails bit a — no named computational result to mirror
Theorem 1 characterizes Luce rationalizability, while the propositions and corollaries characterize experimental-design properties; none supplies a computational result for a population mirror.
I don’t see a qualifying computational anchor in this paper. The closest is Theorem 1, proved here: for any experiment \(\mathcal M\) and stochastic choice dataset \(\theta=(\mathcal M,P)\), \(\theta\) is rationalizable by a censored Luce Choice Rule if and only if, for every overlapping \(C_\theta\)-cycle \([a_1,\ldots,a_{n+1};M_1,\ldots,M_n]\), the products of choice probabilities in the two directions agree:
\[
\prod_{i=1}^{n}P(a_i,M_i)=\prod_{i=1}^{n}P(a_{i+1},M_i).
\]
That is a characterization of rationalizability, not a named result that a computational problem is in P, hard, or otherwise classified. The propositions and corollaries likewise characterize experimental-design properties; the paper gives no computational complexity result to anchor the requested case.
A plausible but unanchored candidate would be a large choice panel whose participants share one of \(\tau\) utility-vector types. Give type \(j\) population mass \(\mu_j\), with \(\sum_j\mu_j=1\). For menu \(M\), aggregate choice shares are
\[
P_\mu(a,M)=\sum_{j=1}^{\tau}\mu_j
\frac{v^j(a)}{\sum_{b\in M}v^j(b)}.
\]
One could ask whether these shares admit a single Luce utility vector \(v\), and return such a vector or a “no” answer. But this changes the paper’s object: it studies a stochastic choice rule, not a distribution of stable utility types, and its observed frequencies are already probabilities. Theorem 1 applies to the resulting shares, but does not establish a complexity result for this population problem. I would not claim that the authors would recognize it as their continuous version, or assign it a complexity class on the strength of this paper.
Theorem 1 is the paper’s strongest possible anchor, and the proponent is right that it is not a computational result: it characterizes which stochastic choice datasets admit a Luce representation. The propositions and corollaries likewise characterize properties of experiments. There is no named complexity problem here whose population version the paper asks us to classify.
The proposed panel mirror adds a distribution over utility vectors, \(\mu_j\), and defines aggregate shares by
\[
P_\mu(a,M)=\sum_j \mu_j\frac{v^j(a)}{\sum_{b\in M}v^j(b)}.
\]
That is a coherent heterogeneity model, but it changes the object Theorem 1 studies. The theorem takes the menu-conditional frequencies \(P(a,M)\) as data and asks whether one utility vector rationalizes them; it neither posits utility types in a population nor retains who made which choices. Many different type distributions can produce the same observed shares on a restricted experiment, and repeating a single type in a larger population leaves the theorem’s input unchanged. Treating the frequencies as population masses therefore continuizes the observed choice probabilities, not the population of the paper’s model.
A stronger version could ask whether a known mixture of Luce types aggregates to shares rationalizable by one Luce utility, or try to recover an unknown mixture. The first adds the mixture as new input to the paper’s existing rationalizability question; the second is a latent-mixture problem requiring structure absent from the paper. Either might be worth studying, but neither is a continuous mirror of a computational result in this paper, and the authors’ interest in it cannot be inferred from their characterization.
So the negative case is strong on the requested anchors: none is a named computational result, and the candidate population model is a substantive extension rather than a high-multiplicity version of the paper’s problem. It is not a proof that no related population-computational project could be worthwhile; that broader claim is too strong to establish from this paper alone.
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.