| paper | d4-journals-jet-2022-0015 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | yes |
Theorem 2 provides a finite test for uniqueness, so it has computational character. But setting \(\mu_\pi=\nu(\pi)\) reproduces the paper’s distribution over rankings, its choice-share equations, and its uniqueness question exactly; the stable-population story changes the interpretation of the masses, not the model or result. This is a restatement of the paper’s existing distributional model, so no distinct continuous mirror survives.
fails bit b-already — the paper is already native to the continuum
The proposed reading covers Theorem 2’s uniqueness test; Theorems 1 and 3 give related graph and support characterizations within the same random utility model.
The strongest honest positive case is Population Preference-Mass Identification, anchored on Theorem 2, which is proved in this paper. One qualification up front: the paper has no named result that states a complexity classification such as membership in P or NP-hardness. Theorem 2 is the closest computational anchor: it gives a finite test for uniqueness. Reading that test as a polynomial-time algorithm requires an explicit representation as input; the paper does not itself state that complexity bound.
The mirror has a natural regime. Imagine a large customer base choosing from a modest set \(X\) of products. Each customer has a stable, complete ranking \(\pi\) of those products, and many customers share each ranking. Let \(\mu_\pi\) be the fraction with ranking \(\pi\), with rational masses and support size \(\tau\). For each offered menu \(A\subseteq X\), experiments on representative cohorts record the fraction choosing each \(x\in A\). These observations satisfy
\[ P_A(x)=\sum_{\pi\in\Pi}\mu_\pi\,\mathbf{1}\{\pi(x)>\pi(A\setminus\{x\})\}. \]
Here \(N\), the number of customers, can be very large while \(\tau\) remains modest. The type is the complete ranking relevant to choice; mass is the population share with that ranking.
The decision problem is: given \(X\), the choice shares \(P_A(x)\) for every nonempty menu, and an explicit rationalizing distribution \(\mu\) over its supported rankings, determine whether \(\mu\) is the only distribution over rankings that rationalizes those shares. Equivalently, ask whether
\[ \{\lambda\in\Delta(\Pi): P_A(x)=\sum_{\pi\in\Pi}\lambda_\pi\mathbf{1}\{\pi(x)>\pi(A\setminus\{x\})\}\ \text{for every }A,x\} =\{\mu\}. \]
The answer is yes or no; in a no instance, another rationalizing distribution \(\lambda\ne\mu\) witnesses non-uniqueness. This is an identification question, so it has no optimization objective.
Theorem 2 states that, for a rationalizable system of choice probabilities, its rationalizing distribution \(\nu\) is unique if and only if there are no two supported orders \(\pi,\pi'\) and alternatives \(x,y,z\) such that: both orders rank \(x\) and \(y\) above \(z\); \(x\ne y\); their weak upper contour sets of \(z\) differ, \(U_\pi(z)\ne U_{\pi'}(z)\); and \(U_\pi(x)=U_{\pi'}(y)\). This is the paper’s own characterization, not a result imported from elsewhere. Given the explicit support, one can check the condition by enumerating pairs of supported rankings and triples of alternatives, so the test is polynomial in \(m\) and \(\tau\). That runtime is a direct consequence of the stated test, rather than a complexity theorem asserted by the authors. The expected direction for this mirror is therefore tractability for the uniqueness decision with an explicit rationalizing distribution. Theorem 2 alone does not settle the complexity of finding such a distribution from choice data when none is supplied.
This is a plausible population reading of the authors’ problem because the paper’s observables are precisely aggregate choice shares, and its unknown is precisely a distribution over complete preference types. For these marginals, a population with fixed preferences and an individual who draws a preference at each choice generate the same equations. The mirror covers Theorem 2’s uniqueness question, not the paper’s other results or broader random utility literature. It also suggests extensions to incomplete menus, noisy or finite-sample choice shares, and recovering the full set of rationalizing distributions.
The soft spot is substantial: the paper already represents the unknown as a probability distribution over preferences. The proposed mirror may therefore be a population interpretation of the paper’s existing model, rather than a distinct continuization. The interpretation is strongest for the paper’s marginal choice data; it would not automatically extend to questions about within-person dependence across repeated choices. I would defend this as an exact, recognizable high-multiplicity reading of Theorem 2, while conceding that under ChoCo’s novelty test it may count as a collision rather than a new continuous mirror.
I’d reject the proposed mirror as a ChoCo contribution, though not because Theorem 2 lacks computational content. Given an explicit support, its condition yields a polynomial-time test: check pairs of supported rankings and triples of alternatives, comparing their upper contour sets. The paper does not state a complexity classification, but the proposed yes/no problem is already answered by its theorem.
More fundamentally, the proposed society is the paper’s object under a change of interpretation. Set \(T=\Pi\) and \(\mu_\pi=\nu(\pi)\). Then the population share choosing \(x\) from menu \(A\) is exactly
\[
P_A(x)=\sum_{\pi\in\Pi}\mu_\pi\,\mathbf{1}\{\pi(x)>\pi(A\setminus\{x\})\}.
\]
The paper studies whether this distribution over complete preference types is identified from those aggregate choice probabilities. Its introduction even describes the unknown preferences as types in a population. A population with fixed ranking types and a RUM agent drawing rankings anew may differ in cross-menu dependence, but the paper observes and uses only the marginal choice probabilities \(P_A(x)\). For the uniqueness question, that distinction changes none of the data or the theorem. Requiring rational masses or a small support only restricts the instances.
The other results do not open a separate mirror: Theorem 1 characterizes uniqueness through the same mass flows on ranking paths, and Theorem 3 links uniqueness of the distribution to identification of its support. Even a computational version taking the full choice table as input has a direct route through Theorem 1: compute the BM-polynomial edge weights, then use positive-edge reachability in the subset graph to detect branching. That is polynomial in the size of the explicit table, though the table itself has exponentially many menus in \(m\).
A new task such as changing population masses to induce identification, or using panel data to distinguish persistent preferences from redraws, could be worthwhile. It would add an intervention or new observations that the paper’s results do not address; it is not a stronger continuous mirror of the anchored uniqueness result. The negative case is strongest for the direct mirror: the model and question are already distribution-over-types identification. I cannot rule out every adjacent research question, but none of the proposed anchors survives as a distinct continuization.
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.