| paper | d4-conf-aldt-2017-0015 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | medium |
| authors would recognise it | unclear |
Theorem 2 is a computational result, and the large-course limit is a plausible extension of the paper’s one-shot equilibrium question. But the proposed finite types do not clearly preserve the global turn positions that determine an individual student’s incentives. The cases leave unresolved whether those incentives admit a compact, convergent population model.
fails bit b-changes — the continuum limit changes the question
Students grouped by preferences and a finite role may still occupy different global turns, so a type-level flow may not determine any one student’s remaining seats or unilateral payoff.
fatal: False
It is undecided whether a compact finite description of types and schedule positions can preserve the residual stock relevant to a tagged student’s deviation as \(q\to\infty\). A convergence and compression proof would settle this in favor of the mirror; showing that exact positions require \(\tau\) to grow with \(q\), or that only a different limit equilibrium remains, would settle it against.
The proposed mirror covers only Theorem 2’s bluff-profile construction and its equilibrium guarantee; it leaves Theorem 1’s cycling result, Theorems 3–5 on crossout and Pareto properties, and Theorem 6 on Stackelberg optimization untouched.
The strongest positive case is a narrow mirror of the paper’s bluff-profile result. It is a candidate for large course-allocation systems where the number of students is far larger than the number of preference and utility types. Seat supply scales with enrollment, and a type records a student’s complete preferences, utilities, and turn role. Mass represents the fraction of students of each type. Individual students still receive indivisible seats; only the type-level allocation is fractional.
The anchor is Theorem 2, proved in this paper. Section 4 gives a linear-time construction: simulate sequential allocation using truthful rankings, then have every agent report the order in which items were picked. The resulting profile has the same pick order and allocation as truthful play, and is a pure Nash equilibrium for every cardinal-utility profile consistent with the ordinal preferences. The proof uses Lemma 3, which invokes the earlier one-other-agent result of Bouveret and Lang [6]. This is the best anchor for a population mirror because it covers an unbounded number of agents. The paper’s other named computational algorithm, Theorem 6, is restricted to two agents. The cited NP-hardness results in the introduction are prior work, not hardness theorems proved here.
I would put this problem to a prover:
Continuous Bluff-Equilibrium Search. An instance gives a finite set \(J\) of seat classes, rational per-capita supplies \(a_j\), a finite set \(T\) of student types with rational masses \(\mu_t\) summing to \(1\), and a finite picking template. Each type specifies a strict ranking and positive additive utilities over the labelled seats, along with its role in the template. A fixed rule orders copies within each seat class. The template and supplies are chosen so that each student has the same finite number of turns and total supply matches total turns per capita.
Interpret this as the limit of finite clone instances: for each common denominator scale \(q\), create \(q\mu_t\) students of type \(t\) and \(qa_j\) seats of class \(j\), expanding the picking template accordingly. The task is to compute the limiting common report order \(\beta\) and type-level allocation flow \(x\) obtained by the bluff construction. A solution must show that reporting \(\beta\) reproduces the truthful limiting allocation and that no individual’s unilateral report change yields a positive limiting utility gain. Seat copies remain indivisible for each student; \(x\) records only what fraction of a type receives each class of seat. The desired running time is polynomial in the number of seat classes, the number of represented types, the template size, and the binary input length \(L\), rather than in the clone count \(q\).
This preserves the paper’s underlying question—computing a Nash equilibrium of a one-shot sequential-allocation reporting game—and its special contribution, a common report that is an equilibrium while reproducing the truthful outcome. The large-university regime gives a credible high-multiplicity interpretation: \(N\) students can be much larger than \(\tau\), the number of distinct complete types. Scaling seat supply with the population is the necessary co-modification; fractional type-level averages describe bundles of real seats rather than fractional seats assigned to an individual.
I expect this narrow search problem to be Class A, because the candidate equilibrium is built from the truthful allocation and its induced item order. That is a prediction, not a result: a compact way to compute the limiting order and flow, and to certify individual deviations at supply-exhaustion points, still needs to be proved. It raises follow-up questions about convergence of the clone profiles, exact complexity in \((|J|,\tau,L)\), and whether the cited hardness of computing a best response disappears in the fluid limit.
The weakest point is the limit of the incentive condition. A student can be pivotal when a finite seat supply is nearly exhausted; one student has zero mass in the continuum. Defining the mirror through limiting unilateral gains keeps the deviation individual, but the equilibrium property need not pass automatically through the limit. If making the model precise requires replacing individual deviations with whole-cohort deviations, the mirror becomes substantially less faithful. My case covers Theorem 2’s bluff equilibrium only; it does not extend to the paper’s other results.
Theorem 2 is a genuine computational anchor: it gives a linear-time method for finding a pure Nash equilibrium, so the “no named computational result” objection does not apply. The problem is that sequential position is part of what makes an agent a distinct strategic player.
In a repeated picking template, two students with the same preferences, utilities, and within-template role still act at different points in the global sequence. The available items at those points can differ, so their roles are not complete types for this game. Recording each student’s actual turn positions fixes that omission, but in an ordinary sequential schedule it gives essentially one type per student. Then \(\tau\) grows with the population, and the proposed high-multiplicity compression disappears.
A stronger mirror could make normalized schedule position part of a continuous type and ask for a limit equilibrium. That raises a more basic issue: the population flow records aggregate seat use, while Nash equilibrium tests each person’s unilateral report. A student takes only a finite number of seats, so their effect on the aggregate is zero, while the utility from their own seats remains finite. At supply-exhaustion points, a tagged student’s payoff can depend on the exact residual seats and turn order—information a mass flow may discard. An almost-everywhere equilibrium could sidestep exceptional positions, but it would change the paper’s all-agents PNE guarantee. Keeping exact individual incentives calls for a tagged-player model with schedule and residual-stock detail, which weakens the claimed compression.
There is also little room for a new complexity landscape unless that compression works. The bluff construction simulates the truthful sequence and returns its item order; its running time is already linear in the explicit instance. Scaling seat copies makes both the schedule and that output grow with the population. A class-level limit could compress them, but it must preserve the turn-position information needed to check deviations. Without that, it computes a fluid allocation, not the equilibrium problem in Theorem 2.
So I would reject the proposed mirror as currently specified: “role in the template” does not make repeated students strategically identical. The negative case is not airtight. A carefully defined continuum of schedule positions with tagged-player incentives might be worthwhile, especially for course allocation, but the paper and the proponent’s formulation do not yet provide a compact model that preserves those incentives.
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.