| paper | d4-conf-sagt-2019-0024 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | no |
Theorem 1 is a computational result, but its proposed continuous version does not preserve the problem it analyzes. In the atomless model, one person’s nomination has no effect on the aggregate, so impartiality becomes automatic; the stratum averages also cannot distinguish profiles with very different individual maximum in-degrees. No continuous question survives that preserves the paper’s individual selection objective and impartiality constraint.
fails bit b-changes — the continuum limit changes the question
The attempted mirror covers Theorem 1’s Sample and Vote guarantee; it leaves Theorem 2, Theorem 3, Corollaries 1–2, and Theorem 4 untouched.
The strongest positive case I can make is a narrow, conditional mirror of Theorem 1, proved in this paper. It says that in the single-nomination model, where every agent nominates exactly one other agent, the randomized Sample and Vote mechanism is impartial and has additive error \(\Theta(\sqrt n)\) when its sample size is \(k=\Theta(\sqrt n)\). The paper’s other main results concern mechanism guarantees and impossibility bounds; it has no named theorem classifying a problem as NP-hard, in P, FPT, or similar. So Theorem 1 is an algorithmic anchor, not a complexity-class result.
A plausible large-community setting is peer recognition in a professional community with many members but relatively few relevant strata, such as role and specialty. Let \(\mathcal R\) be those strata. A type \((r,s)\) consists of an agent in stratum \(r\) who nominates a uniformly chosen member of stratum \(s\). The society is a distribution \(\mu_{r,s}\) over these types, with \(\sum_{r,s}\mu_{r,s}=1\); its support size \(\tau\) counts represented types, and may be much smaller than the population. Write \(\bar\mu_r=\sum_s\mu_{r,s}\). The expected nomination rate per member of stratum \(r\) is
\[
d_r(\mu)=\frac{\sum_s\mu_{s,r}}{\bar\mu_r}
\]
when \(\bar\mu_r>0\). These rates can be fractional even though each individual casts one nomination, the sort of type-level averaging the programme allows.
Here is the continuous problem I would hand to a prover, as a mirror of Theorem 1: given a finite stratum-and-action schema \(\mathcal R\) and \(\varepsilon\ge 0\), does there exist an anonymous measurable selection rule \(q\), defined for every atomless society \(\mu\) on that schema, such that \(\sum_r\bar\mu_r q_r(\mu)=1\), the selection density for a member of stratum \(r\) is independent of that member’s own nomination, and
\[
\max_r d_r(\mu)-\sum_r\bar\mu_r q_r(\mu)d_r(\mu)\le\varepsilon
\]
for every \(\mu\)? A solution is the rule \(q\). This is the normalized additive-efficiency question behind the paper’s single-nomination result. It is my lead candidate, though I would describe it as a limit question rather than claim it faithfully retains the paper’s full problem.
Under this formulation the answer is immediately \(\varepsilon=0\): select uniformly from the strata with maximum nomination rate, then uniformly within the selected stratum. In an atomless population, one person’s changed nomination has zero effect on the aggregate distribution, so it cannot change their selection density. The question is tractable, but that tractability may be an artefact of the limit rather than a continuization result of the kind ChoCo seeks. It generates a more substantive follow-up: which finite-\(n\) conditions recover a nontrivial impartiality constraint and the approximation rates in Theorem 1?
That is the weak point, and it is serious. In the paper, one person’s nomination can change another candidate’s score by one and can affect the winner’s probability. In the proposed mirror, individual deviations are invisible. Treating a whole type’s mass as the deviating actor would restore influence, but would turn a type into a coalition and change the problem. Also, the nominees are named individuals; grouping nominators into few types does not by itself make the selection target a few types. An author might accept this as a mean-field benchmark for the normalized guarantee, but could reasonably reject it as a mirror of impartial selection itself.
The proponent has one real anchor: Theorem 1 gives an explicit randomized algorithm and an additive guarantee. I would not dismiss it as having no algorithmic content. But it does not classify the complexity of a computational problem; its result is a guarantee for a rule on finite nomination graphs. The proposed continuous question asks whether a rule \(q\) exists, with no input-wise computation to analyze. Making that question computational requires a finite description of the society—and that is where the proposed mirror loses the paper’s central objects.
A type \((r,s)\) records a nominator’s stratum and the stratum they nominate, not the identity of the nominee. That aggregation can equate profiles whose quality differs by almost \(n\). With one stratum, a directed cycle and a profile where everyone nominates the same person both have \(\mu_{r,r}=1\) and \(d_r=1\); their maximum individual in-degrees are \(1\) and \(n-1\). The proposed benchmark, the best stratum-average rate, cannot distinguish them. Selecting a random member of the stratum with the highest rate may be a sensible group-level problem, but it is not the paper’s problem of selecting a high-in-degree individual. The difficulty here is not that the continuous answer is easy; the aggregate input no longer represents the instances the theorem measures.
A stronger mirror might keep each person as a point in an atomless population \(X\), record nominations as a map \(f:X\to X\), and choose a probability measure over winners. But a society represented by aggregate measures cannot distinguish two nomination maps that differ at only one person. Changing that person’s nomination therefore leaves the continuous state unchanged, making their own-action constraint automatic for any rule based on that state. And if nominations are represented as ordinary mass, a single nomination has zero mass: the finite in-degrees and sampling effects that drive the paper’s additive guarantees disappear. Keeping those effects requires atomic individuals or identity-indexed incidence data. The former brings back the finite-agent model; the latter restores the individual graph rather than compressing it into a small-support society.
The stratum version has the same problem in another form. If a type’s mass represents many people who nominate a target *class*, then choosing a person uniformly from the winning class changes the benchmark to class-average nominations. If instead the type records a particular target’s identity, the representation must retain those identities, and one-person deviations again have no effect on aggregate mass. Treating a positive mass of a type as the deviating actor would make impartiality constrain a coalition, not an individual.
The scale reinforces the mismatch: normalizing Theorem 1’s \(\Theta(\sqrt n)\) additive error by population size gives \(\Theta(1/\sqrt n)\), which vanishes. The continuous limit can express normalized nomination rates, but not the finite-scale distinctions the theorem bounds. Retaining those distinctions calls for a finite-\(n\) or identity-sensitive model, rather than a continuous-population computational problem.
This is a strong case against Theorem 1 as a ChoCo mirror, not a proof that no adjacent continuous study could be worthwhile. A group-level selection problem or a sparse-network limit could be interesting in its own right. But each either changes the paper’s benchmark or retains the individual structure that the continuous mirror was meant to replace.
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.