| paper | d4-conf-ecai-2024-0298 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | yes |
The article has no numbered computational result: its Strong-PJR statement is a representation guarantee, and its experimental observations have no stated computational guarantee. The proposed continuous feasibility problem is well-posed and related to the article’s model, but it is a new computational question rather than a mirror of a computational result in this paper.
fails bit a — no named computational result to mirror
The article’s Strong-PJR statement is not a named computational result, so the proposed feasibility problem has no computational anchor in this paper.
fatal: True
No computational result in the article is covered; it discusses proportional aggregation applications, states a Strong-PJR guarantee, and reports preliminary experimental observations.
The strongest positive setting is repeated evaluation in RLHF: a very large pool of labelers sees many prompts, and a relatively small number of stable preference types recur across prompts. A type is one complete approval pattern over the response options in every round; its mass is its fraction of the labeler pool. The decision is one response per prompt. This is close to the article’s sequential approval model, and preserving labeler identity across prompts is central to the article’s motivation.
A concrete candidate is Continuous Strong-PJR realization. An instance has rounds \(j=1,\ldots,T\), alternatives \(C_j\), and types \(t=1,\ldots,\tau\), with approval sets \(A_{tj}\subseteq C_j\) and rational masses \(\mu_t\) summing to \(1\). For a nonempty type set \(S\), let \(\mu(S)=\sum_{t\in S}\mu_t\), and let \(k(S)\) be the number of rounds in which the types in \(S\) have a common approved alternative. Find a sequence \(D=(d_1,\ldots,d_T)\), \(d_j\in C_j\), such that for every \(S\),
\[
\left|\left\{j:d_j\in\bigcup_{t\in S}A_{tj}\right\}\right|
\ge \left\lfloor\mu(S)k(S)\right\rfloor.
\]
This uses the paper’s “happy” condition: a coalition is happy in a round if at least one of its members approves the chosen response. The floor is the natural discrete quota: in a finite electorate with mass \(\mu(S)=|S|/n\), it reproduces the Strong-PJR guarantee.
Here is the anchor limitation: the supplied article has no numbered theorem, lemma, proposition, or corollary asserting a complexity result. It reports that Phragmén’s rule satisfies Strong PJR and describes experiments as preliminary evidence; neither is a numbered complexity anchor in this article. Its technical companion has Theorem 4.1, “Sequential Phragmén satisfies Strong PJR,” but that result is in Chandak, Goel, and Peters’s separate AAAI paper, not this article, and it is a proportionality guarantee rather than a complexity classification. [The companion paper](https://ojs.aaai.org/index.php/AAAI/article/view/28813/29553)
If that external theorem is admitted as context, the candidate problem looks like a plausible Class A question: can a weighted, type-compressed version of the sequential rule produce such a sequence in time polynomial in \(T,\tau\), the input size of the alternatives, and the mass bit length \(L\)? Expanding rational masses into individual voters may require a population size exponential in \(L\), so avoiding that expansion is the substantive algorithmic question. The authors would likely recognize the model, but this proposed complexity question is not an anchored result of the supplied article.
The weak point is the regime: labelers must share their complete approval behavior across the relevant prompts. If preference patterns proliferate with the number of prompts, then \(\tau\) may approach the number of labelers and the high-multiplicity advantage disappears. The mirror is credible for stable preference cohorts, but the article itself does not establish that computational case.
The strongest negative is about the paper-to-mirror link. This article states no named computational result: its formal claim is that Phragmén satisfies Strong PJR, alongside preliminary experimental observations. The proponent’s “Continuous Strong-PJR realization” instead takes a theorem from a separate paper and turns an axiomatic guarantee into a new feasibility problem. That companion theorem establishes a representation property, not a complexity classification. So the proposed problem may be inspired by this article, but it is not a continuous mirror of a computational result in it.
The best version of that proposal is still plausible: imagine a large feedback panel with stable preference types answering a fixed set of prompts. A distribution over complete approval trajectories would preserve who agrees across rounds, and the continuous Strong-PJR question would be well-posed. But that plausibility does not supply the missing computational anchor. In open-ended RLHF, complete trajectories across prompts may produce nearly one type per labeler; compressing them into a few latent archetypes would require an additional modeling or inference step. The article identifies sparse, identity-linked feedback as a challenge, not a computational problem it analyzes. Its LLM and virtual-democracy examples likewise motivate proportional aggregation without stating a complexity question to mirror.
This defeats the paper-specific anchor, but it does not support the universal claim that no worthwhile continuous project exists. The stable-cohort version could be a sensible high-multiplicity problem in its own right, and the supplied material gives no honest basis for ruling out every such scenario. The negative case is strong only in the narrower conclusion: this article motivates continuous computational work, but does not itself provide a computational result for that work 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.