| paper | d4-journals-geb-2025-0097 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | medium |
| authors would recognise it | no |
Theorem 3.1 is a named computational performance result: it guarantees rank outcomes for a stated algorithm under the paper’s random model. But the proposed continuum version replaces rankings over named individuals with rankings over counterpart types, and a fixed positive mass gap does not preserve the theorem’s one-doctor surplus. That is a different enough matching process that no continuous question survives as a mirror of this result.
fails bit b-changes — the continuum limit changes the question
The attempted mirror concerns only Theorem 3.1’s rank outcomes under doctor-proposing deferred acceptance; it does not cover the proof technique or the paper’s comparisons with balanced markets.
The strongest honest positive case is a near miss, not a qualifying mirror. The paper has no named result asserting a complexity classification such as polynomial time or NP-hardness.
The closest anchor is Theorem 3.1, proved here: with \(n+1\) doctors and \(n\) hospitals whose complete preference rankings are sampled independently and uniformly, doctor-proposing deferred acceptance produces a stable matching \(\mu\) such that, asymptotically almost surely, hospitals’ average rank of their matched doctors is \(\Theta(\log n)\), while doctors’ average rank of their matched hospitals is \(\Theta(n/\log n)\). The short-side phenomenon was established in prior work; this paper’s contribution is a direct proof using the doctor-proposing algorithm. The theorem is a probabilistic performance result, not a hardness or tractability result. The paper’s lemmas support that analysis but do not supply a qualifying computational anchor either.
A neighboring continuous question could take types to be complete rankings over counterpart types, assign each type a rational mass, and ask for the mass-matching produced by a type-level deferred-acceptance process and its average ranks. But this is difficult to defend as a mirror of Theorem 3.1. In the paper’s random model, rankings are over named individuals and are overwhelmingly distinct, so the natural support size grows with the population. Replacing individuals with counterpart types changes who agents rank and how proposals work. Also, the market’s one-doctor surplus has vanishing mass as \(n\) grows; a continuum model with a positive supply gap is a different regime, while an equal-mass limit loses that surplus.
So I would not assign this a Class A, B, or C expectation: there is no anchored computational problem here whose complexity the continuous version could resolve. The paper motivates studying large matching markets, but a continuous type-level deferred-acceptance problem would be new research rather than a continuization of one of its named complexity results.
Theorem 3.1 is the only plausible anchor, and it fails the programme’s computational test. It proves a high probability rank law for doctor proposing deferred acceptance under a random profile generator. It poses no decision or optimization problem and gives no complexity classification or algorithm for a computational task. The supporting lemmas bound events in that random process; turning them into a request for an algorithm would require adding a task the paper does not study.
The strongest repair would use finitely many preference types with rational masses and ask for an algorithm to compute a type level deferred acceptance outcome or its rank statistics. The paper’s preferences, however, are rankings over named individuals, and the complete type must capture the information that determines how those individuals rank one another. With independent uniform rankings, those complete types are essentially singletons, so the support grows with the market. Collapsing people into cohorts makes a credible high multiplicity matching model, but changes rankings to be over counterpart types and replaces the individual proposal process with a mass process. That may be a worthwhile problem in its own right; it is not a computational version of the paper’s result.
The one extra doctor does create a vanishing mass gap, but that objection can be repaired by studying a market with a fixed positive supply gap. Even with that repair, the missing computational task and the identity level random rankings remain. The paper’s result is about the typical performance of a particular algorithm under a particular random model, not about computing an outcome from a distribution over repeated types.
So the negative case is strong for this paper as a continuization anchor, though it cannot rule out continuous computational work on type based matching more broadly. Such work could be sensible for ChoCo; the paper offers motivation for it, but no named computational result to mirror.
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.