| paper | d4-journals-geb-2026-0010 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | yes |
The paper defines the JHM fitting objective and reports using brute force, but it has no named theorem, lemma, corollary, or proposition asserting an algorithmic or complexity result. The proposed population version is statable, but it is unanchored and cannot pass the computational-result requirement.
fails bit a — no named computational result to mirror
The proposed aggregation extends the individual JHM fitting and preference recovery in Sections 2.3 and 4.1; it leaves the experimental design and the paper's empirical findings untouched.
This paper has no qualifying named computational result to anchor a continuous mirror. It contains no numbered Theorem, Lemma, Corollary, or Proposition asserting a complexity or algorithmic result. The numbered theorems it mentions are results by other authors, and the paper’s own computational claims—defining JHM, using brute force, and reporting that the computation takes under 12 seconds for six alternatives—are not numbered results of that kind. So I cannot honestly present an anchored case as requested.
The strongest plausible mirror suggested by the paper’s method is a population version of JHM fitting. Imagine a very large panel answering the same finite collection of choice menus. Its input is the rational mass \(\mu_d\) of respondents with each full observed response record \(d\), including the treatment and any other measured features used in the analysis. The support size \(\tau\) is the number of distinct records, which could be much smaller than the number of respondents in a repeated, standardized survey. The model and preference relation that best explain each record remain to be inferred.
Here is a precise candidate, though it is not anchored by a named result in the paper. Let \(X\) be the alternatives, \(\mathcal A\) the menus, and \(d\) a record giving a chosen set \(C_d(A)\) for each menu \(A\in\mathcal A\). For the forced-choice setting, let \(R\) contain the paper’s utility-maximization and undominated-choice model/preference pairs, with predicted choice set \(C_r(A)\). Define
\[
\delta(d,r)=\sum_{A\in\mathcal A}
\left(1-\frac{|C_d(A)\cap C_r(A)|}{|C_d(A)\cup C_r(A)|}\right).
\]
Choose masses \(x_{d,r}\geq 0\) that assign each response type to explanatory model/preference pairs, with \(\sum_r x_{d,r}=\mu_d\), to minimize
\[
\sum_{d,r}x_{d,r}\,\delta(d,r).
\]
A solution is the minimizing assignment and its total JHM discrepancy. This is the mass-weighted version of the paper’s individual fitting objective; the latent explanations are preferences and choice models, not fractional outcomes. It is an LP with potentially many explanation variables, and its computational bottleneck is finding a best-fitting relation for a response record.
The case for author recognition is that this preserves the paper’s choice models and its specific Jaccard penalty for partial versus full choice mismatches. A panel-level distribution over response records is plausible when many respondents face the same menus, but this formulation is an extension of the paper’s individual analysis, not one of its stated results. I would expect no new tractability from the population step: the instance with all mass on one record is already the individual JHM problem. If that underlying problem is hard as the number of alternatives grows, the hardness carries over; the paper itself establishes neither that hardness nor a polynomial algorithm. The natural further question is whether the best-fitting preference can be found efficiently, or approximated, as a function of \(m\), \(\tau\), the number of menus, and the input bit length.
This mirror covers only the JHM-based recovery of models and preferences. It does not mirror the experimental incentive design or the paper’s statistical findings. Its weakest point is that mass-weighted fitting may amount to batching the same independent individual analysis, rather than revealing a new population-level computational problem. Also, if most respondents have distinct response records, then \(\tau\) may be close to the panel size and the high-multiplicity premise offers little advantage.
The main obstacle is that the paper has no named computational result to continuize. Its JHM index is a definition, and its brute-force implementation takes under 12 seconds on the six-alternative experiment. Neither is a theorem about an algorithm’s complexity. The paper’s own numbered computational result, as the other analyst acknowledges, is absent. Their proposed LP is therefore a new question suggested by the paper, not a mirror anchored on one of its results.
Grant the strongest plausible setting for that proposal: a large consumer panel sees the same menus, and many respondents have identical full response records. Treating each record \(d\), including its treatment and any relevant measurements, as a type is a sensible high-multiplicity model. But the proposed objective separates by record. If \(R\) is the set of model and preference explanations and \(\delta(d,r)\) is the JHM discrepancy, then
\[ \min_{x\geq 0}\sum_{d,r}x_{d,r}\delta(d,r) \quad\text{subject to}\quad \sum_r x_{d,r}=\mu_d \]
reduces to \(\sum_d \mu_d\min_{r\in R}\delta(d,r)\). Each record gets its best explanation independently; the population mass only weights the resulting scores. Repeated records can be processed once and counted by their mass, but the population introduces no shared constraint or collective choice to optimize. The paper’s reported computation could motivate work on the individual fitting problem as the number of alternatives grows. That would be a computational question, but the paper states no complexity result for it to mirror, and the population formulation adds only aggregation.
A more ambitious version would pool menu-level choice proportions and ask whether they can be explained by a distribution over preference relations. That could be a legitimate continuous-population problem, but it loses the links between the choices made by the same respondent across menus. Those links are what let the paper recover individual relations and distinguish indifference from incomparability or indecisiveness. Impose a common relation to restore a single explanation, and the model no longer represents the heterogeneity the paper studies. Retain full response histories to preserve that target, and the fitting again separates by history. A noisy latent-type model might create a richer joint inference problem, but it adds a stochastic response model and changes the paper’s deterministic, respondent-level task.
The paper’s introductory policy examples make population aggregation imaginable, and repeated consumer panels make the high-multiplicity regime plausible. So I would not reject this mirror on the ground that a continuous society is incoherent, or that its answer is already known. The negative case is narrower: the faithful continuous version is a weighted batch of individual fits, while the more coupled population questions require changing what is being inferred. I would advise against giving this paper a continuization slot on the strength of the proposed LP. The residual weakness is that a newly specified population inference problem could still be worthwhile; it would simply lack an anchor in the paper’s named computational results.
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.