| paper | d4-conf-aaai-2014-0033 |
| 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 characterizes statistical accuracy under noise, while Theorems 2 and 3 concern the same robustness property; none gives an algorithm or complexity result. A distribution over ranking types is natural here, but replacing noisy samples with exact masses removes the accuracy-in-the-limit question, and computational questions about recognizing robustness would require a new problem formulation.
fails bit a — no named computational result to mirror
No named result yields a computational mirror. Theorems 1–3 address statistical robustness and properties of voting-rule families.
I can’t give a qualifying anchor for a continuous computational mirror: the paper contains no named result asserting a problem is tractable, hard, or otherwise classified by computational complexity.
The nearest result is Theorem 1, proved here: among generalized scoring rules without holes, a rule is monotone-robust with respect to every distance metric if and only if, on every profile with a unique most frequent ranking, it outputs that ranking with probability \(1\). That is a characterization of statistical robustness, not a computational result.
There is a natural continuous-population description close to the paper’s own model: rankings are types, \(\mu_\sigma\) is the fraction of the crowd reporting ranking \(\sigma\), and the aggregation rule maps that distribution to an output ranking. But the paper’s anchors concern accuracy under increasingly many i.i.d. noisy votes. They do not study the computational complexity of a problem over population masses. Turning that description into a ChoCo-style prover question would therefore require adding a new computational problem, rather than mirroring a named computational result of this paper.
The strongest negative is that this paper has no named computational result to mirror. Theorem 1 characterizes which generalized scoring rules are statistically robust to every metric based noise model. Theorems 2 and 3 concern the same robustness property: which familiar rules satisfy it, or cannot satisfy it. None classifies the complexity of an algorithmic problem.
The population itself is a perfectly sensible high multiplicity object here: rankings are types, and their masses form \(\mu\). But the paper already represents a profile by its empirical distribution \(x^\pi \in \mathbb{Q}^{m!}\), and its noise models are distributions over rankings. Recasting profiles as population masses does not create a new computational task. If we ask for the continuous rule’s output on \(\mu\), the robust rule identified by the theorem selects its most frequent ranking. If we ask whether a rule is robust for every \(\mu\) and every permitted noise model, we have restated the theorem’s global statistical property, not obtained a complexity problem over population instances.
A more ambitious proposal could ask how hard it is to recognize robustness from a succinct description of a rule, or to optimize over noise models. Those might be worthwhile computational questions. But their difficulty would depend on how rules and noise models are encoded, and their input is the rule description or noise model—not the size of the crowd or the support of its type distribution. They would be new questions, rather than continuous mirrors of a computational result in this paper.
The case against is strongest on scope, not on whether crowdsourcing admits a continuum model. High multiplicity is natural in the paper’s setting, and the continuum is already native to its statistical formulation. The limit deliberately studies recovery as sampling grows; replacing samples with exact masses removes the sampling process that defines accuracy in the limit. That does not rule out useful computational work on crowdsourcing robustness, but it leaves no convincing ChoCo mirror among this paper’s results. The universal claim that no such scenario could ever be worthwhile is harder to defend; the narrower claim that this paper supplies no computational result for continuization is solid.
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.