| paper | d4-conf-ecai-2023-0164 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | yes |
Lemma 1 is the only numbered formal result the proponent identifies from this paper, and it makes a geometric claim rather than a computational one. The recognition algorithm and weighted subprofile experiments are not stated as a qualifying numbered computational result, while the cited NP-hardness result belongs to Peters, not this paper.
fails bit a — no named computational result to mirror
The proposed mirror extends Section 5.1.2's voter-deletion heuristic to explicit type masses; it leaves the recognition algorithm and Lemma 1 without a qualifying computational theorem.
The strongest honest mirror is Continuous 2-Euclidean Recognition. There is one qualification up front: the paper has no numbered theorem, lemma, corollary, or proposition asserting a complexity classification for its recognition problem. Its main contribution is Algorithm 1, an incomplete recognition algorithm evaluated experimentally. The introduction reports an external hardness result, but prints no theorem number for it. So I can give a concrete mirror, but not a qualifying named anchor under your rule.
For the mirror, an instance is a candidate set \(C\) and a finite set \(T\) of strict rankings, each with positive rational mass \(\mu_t\), where \(\sum_{t\in T}\mu_t=1\). A type is a complete ranking; mass is the fraction of the electorate with that ranking. Ask whether there are candidate locations \(p_c\in\mathbb R^2\) and, for each type \(t=(c_{t,1},\ldots,c_{t,m})\), a voter location \(q_t\in\mathbb R^2\) such that
\[
\|q_t-p_{c_{t,1}}\|_2 < \cdots < \|q_t-p_{c_{t,m}}\|_2.
\]
A YES solution is such a placement; a NO answer means no placement exists. This is a feasibility question, not an optimization problem. Each mass \(\mu_t\) can be placed at \(q_t\).
The natural regime is a large electorate with many repeated rankings: \(n\gg\tau\), where \(\tau=|T|\le m!\). Ranking is a complete type for this problem because the paper’s model uses preferences alone. The authors would likely recognize the question: their recognition task already ignores duplicate preferences, so replacing each distinct ranking by its population mass preserves exactly the input distinction their algorithm uses. The mirror covers recognition and its YES/NO object, not the algorithm’s timeout behavior or its random search over candidate locations.
The paper’s closest formal structural result is Lemma 1, proved here. It says that if one candidate’s location is a convex combination of other candidates’ locations, then for every voter position inducing a strict ranking, that candidate is preferred to at least one of those other candidates; in particular, it cannot be ranked last. In the mirror, any type with positive mass ranking that candidate last therefore forces its location to be a vertex of the candidates’ convex hull. This helps explain the paper’s candidate-position generation, but it is not a complexity anchor.
The paper also reports that Peters [29] proved recognition under the Euclidean \(\ell_2\) norm NP-hard for every fixed dimension \(d\ge2\), and that some profiles require exponentially many bits to specify any embedding. That is relevant outside support, but it is not a numbered result of this paper. If used to predict the mirror’s classification, the answer is hardness transfers: a finite profile can be encoded by its support with, for example, equal positive masses, and recognition depends only on that support. The masses do not dissolve the hardness.
A more population-sensitive follow-up is to choose candidate locations maximizing the mass of rankings they can explain:
\[
\max_{p_c\in\mathbb R^2}\ \sum_{t:\,D_t(p)\ne\varnothing}\mu_t,
\]
where \(D_t(p)\) is the set of voter locations inducing ranking \(t\) for the chosen candidate locations. This is a precise weighted version of the voter-deletion approximation studied heuristically in Section 5.1.2, but it has no numbered result to anchor it.
The weakest point is that, for exact recognition, the masses are inert: only whether each \(\mu_t\) is positive matters. This makes the mirror a sensible high-multiplicity formulation and gives a clean hardness-transfer question, but it offers less of the programme’s distinctive opportunity for fractional population mass to change the feasible solutions or unlock a new algorithm.
The strongest case against is decisive on the requested anchors: the paper proves no named complexity result for a computational problem. Its main contribution is an incomplete algorithm with experimental results. The NP-hardness statement is Peters’s result, reported in the introduction, not a result of this paper. Lemma 1 is named, but it is a geometric structural fact, not a computational classification.
Exact continuous 2-Euclidean recognition adds no population question. The paper says duplicate rankings can be removed, and its YES/NO predicate depends only on which rankings occur. For a distribution \(\mu\), recognition is therefore determined entirely by \(\operatorname{supp}(\mu)\); changing the masses while preserving the support changes nothing. Lemma 1 has the same limitation: a ranking with positive mass imposes its geometric condition, but the amount of mass does not matter. These are poor anchors for a programme about computational questions over continuous societies.
The proposed weighted objective,
\[
\max_{p_c\in\mathbb R^2}\ \sum_{t:\,D_t(p)\ne\varnothing}\mu_t,
\]
is the best salvage. It makes mass consequential and is close to the paper’s voter-deletion experiments. But the paper reports a heuristic and measurements for that task, not a named computational result to mirror.
That is a strong case against the proponent’s anchors, but it does not establish the universal claim. The weighted objective is a plausible computational question about continuous population mass, and I cannot honestly dismiss it as unworthy; it may be a useful follow-on rather than a mirror of a named result in this paper. The negative case is strong under the stated anchor requirement and weak against the broader claim that *no* worthwhile continuous question can arise from the paper.
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.