Mechanisms That Play a Game, Not Toss a Coin

The proponent’s only anchor, Theorem 8, is an equilibrium and stability claim rather than a numbered complexity or algorithmic result of the required kind. The paper does contain a numbered approximation theorem, Theorem 4, but no continuous question was stated for it. The proposed atomless DA model is recognizable and potentially Class A, but replacing agent-generated seed play with public \(\omega\) makes it an adjacent extension rather than a mirror of Theorem 8.

The objection that survived

The atomless limit makes each student’s seed and report payoff-irrelevant, so public \(\omega\) replaces the paper’s agent-generated modular randomization with exogenous lottery tie-breaking.

What the mirror covers

The proposal covers only the school-choice application around Theorem 8; it leaves the other domains and results untouched and does not mirror the paper’s cited hardness results.

Open questions for a prover

The case FOR (proponent)

The strongest honest positive case is narrow: a continuous mirror of the school-choice part, anchored on Theorem 8. It should not be presented as a mirror of the whole paper.

Theorem 8 is proved in this paper. It states that the de-randomized deferred-acceptance mechanism has a mixed Nash equilibrium in which at least two students choose uniformly random seed integers and all students choose schools sincerely; the resulting equilibrium induces a distribution over ex post stable matchings.

A natural high-multiplicity regime is public-school choice. A district may have millions of students but a relatively small set \(T\) of complete student types: each type specifies a full school ranking, sibling or district status, and every other characteristic used by school priorities. Let \(\mu_t\) be the fraction of students of type \(t\). Schools \(s\in S\) have capacities \(q_s\), normalized so that \(\sum_s q_s=1\), and priority classes \(r_s(t)\). This is not an artificial replication story: school-choice systems genuinely contain large cohorts of students sharing preference and priority characteristics.

Call the mirror \(\textsc{Continuous-DA-Seed}\). Its input is rational \(\mu\), school capacities \(q\), complete rankings \(\succ_t\), and priority classes \(r_s(t)\). A seed \(\omega\) specifies the tie-breaking order within every school priority class; the continuous analogue of the modular game is a public uniform seed \(\omega\), or equivalently a uniform distribution \(\lambda\) over such seeds. Given \(\omega\), student-proposing deferred acceptance runs on the atomless population. The decision variable is the mass assignment \(x=(x_{t,s})\), where \(x_{t,s}\) is the mass of type \(t\) assigned to school \(s\), together with a measurable realization \(a_\omega\) assigning each infinitesimal student to one school.

The computational question is:

\[ \textsc{Continuous-DA-Seed} \]

Given \((T,S,\mu,q,\succ,r,\omega)\), compute \(a_\omega\) and its aggregate assignment \(x\), or, in the randomized version, compute an \(\varepsilon\)-accurate aggregate assignment under the uniform seed law \(\lambda\). The output must satisfy the row and capacity constraints

\[ \sum_s x_{t,s}=\mu_t,\qquad \sum_t x_{t,s}\le q_s, \]

and a stability certificate: no positive mass of type \(t\) can move to a preferred school \(s\) while displacing only lower-priority mass there. The objective is mechanism evaluation rather than welfare optimization: recover the stable assignment and therefore any linear welfare \(\sum_{t,s}w_{t,s}x_{t,s}\).

This is a credible mirror of Theorem 8 because it preserves the paper’s central objects: deferred acceptance, school capacities, broad priority classes, lottery-based tie-breaking, truthful student reports, and ex post stable assignments. The continuity is in the student population, not merely in the assignment outcome: each infinitesimal student still receives a whole school, while \(x\) records the aggregate mass assigned to each school.

I expect the seed-specific and type-level version to be Class A. With finitely many types, each type can be rejected only finitely many times, so fluid deferred acceptance has polynomially many type-school events and uses rational arithmetic of polynomial encoding length. The exact distribution over all individual priority permutations is a separate question and may be substantially harder; it should not be confused with computing one continuous DA realization. The mirror therefore generates further questions about exact expected assignments, \(\varepsilon\)-approximation, parameterization by \(|T|\), and finite-population rounding.

The weakest point is strategic. In the continuum, an individual student has zero mass and cannot materially change the aggregate seed or assignment. Thus the modular-game incentive that is central to Theorem 8 partly disappears. One could retain a finite seed committee, but that would be a hybrid model rather than a pure population continuum. The positive claim is consequently about the computational and matching content of Theorem 8—continuous stable assignment under endogenous tie-breaking—not a literal preservation of every individual-level strategic feature.

There is also an important qualification about the requested anchors. The paper has no numbered theorem whose conclusion is explicitly “NP-hard,” “in P,” “FPT,” or similar. Its numbered results are equilibrium, approximation, encoding, and impossibility statements. The NP-hardness claims about probabilistic serial and random priority appear only in prose and are cited from earlier papers. Thus, under a strictly complexity-class-only reading of “named computational result,” this paper has no qualifying anchor. Under the broader reading that a named algorithmic equilibrium theorem counts, Theorem 8 is the strongest defensible anchor.

The case AGAINST (opponent, writing after the proponent)

The strongest negative case starts with the paper’s lack of a qualifying anchor. It contains no numbered theorem establishing a complexity classification, an algorithmic running time for a new computational problem, or a hardness result. Theorem 8 is an equilibrium-existence and normative statement. Deferred acceptance itself is standard; the theorem does not analyse the complexity of computing the equilibrium or the induced distribution. The NP-hardness claims concerning probabilistic serial and random priority are cited results from earlier papers, not results of this paper. Theorem 9 is an encoding-length observation, while Theorems 10 and 11 are again equilibrium-existence claims.

Theorem 8 is therefore the only plausible anchor, and its defining object does not survive a pure population continuum. The finite mechanism uses named students in two essential ways: each submits an integer to a modular sum, and the sum selects a permutation of those students. The equilibrium argument relies on an individual’s random seed affecting the common outcome while remaining strategically unpredictable.

In an atomless society, an individual student has zero mass. Changing that student’s seed or school report cannot change the aggregate assignment or any aggregate seed. Consequently, the theorem’s strategic conclusion—sincere school reports together with random seed play—loses its content. Every report is payoff-irrelevant at the margin, so sincerity is no longer selected by the modular game; it is merely one among many indifferent behaviours.

The natural repairs each change the anchor in a material way. A finite committee of seed-generating students preserves the game, but makes the model a hybrid with finitely many pivotal agents rather than a continuous society. Making each type choose a seed turns a type into a collective player, not a mass of interchangeable students. Introducing a public random \(\omega\) restores a lottery over priority orders, but that is randomized deferred acceptance, not the paper’s de-randomization through agents’ play.

There is also a representation problem. A finite uniform permutation has \(n!\) possibilities and can be generated by the inverse Lehmer code. A complete priority order over an atomless population is not a finite object determined by \((T,\mu)\). If \(\omega\) is reduced to an order over types, it no longer represents the paper’s lottery among students within a priority class. If it is represented by independent continuous priority labels, it becomes an exogenous random field rather than the modular game. If only the aggregate assignment \(x_{t,s}\) is retained, the result is a new high-multiplicity deferred-acceptance model, not the ex post stable matching generated by Theorem 8.

That new model may be sensible—large public-school cohorts are a genuine high-multiplicity regime—but it is not a computational mirror of a result in this paper. Seed-specific computation is essentially ordinary deferred acceptance on a compressed population. Expected assignment under random tie-breaking would be a worthwhile high-multiplicity matching problem, but it is an invented extension of randomized DA, not a result the paper establishes or analyses.

The other possible anchors are weaker still. Task allocation is built around exactly two named agents with identity-specific processing times and payments; scaling it to a population changes the mechanism. Peer-selection responsiveness explicitly concerns whether each named individual can change the winner, which becomes vacuous when every individual has measure zero. The resource-allocation results likewise concern finite-agent lotteries and cited complexity results rather than computational theorems proved here.

So I would reject this paper as a source of a ChoCo mirror: there is no named computational result, and the only credible candidate loses its strategic and de-randomizing content under continuization. The limitation is that the universal claim “no worthwhile school-choice continuum exists” would be too strong. A continuous high-multiplicity DA project could be valuable; it simply would be a new project adjacent to this paper, not a faithful mirror of its central result.

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.