| paper | d4-conf-ecai-2023-0092 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | yes |
The paper proposes Embedded Voting and evaluates it experimentally, but contains no numbered computational result of its own; Theorem 4.1 is cited from another source. The proposed mass-weighted EV rule is statable, but it cannot supply the required result-of-this-paper anchor.
fails bit a — no named computational result to mirror
The proposed extension covers EV's candidate-scoring rule; the paper's experimental performance findings and its other aggregation rules have no anchored continuous computational result.
There is a plausible continuous-population angle here, but the paper has no eligible named computational result to anchor it to. It states no numbered theorem, lemma, corollary, or proposition asserting a complexity result. Theorem 4.1 is cited from Kuncheva, not proved here; it concerns the optimality of utilitarian voting under independent Gaussian errors, not computational complexity. The paper’s findings about EV are experimental comparisons. So, under your anchor rule, there are zero anchors and no honest per-anchor questions or complexity classifications to report.
The strongest unanchored mirror I can see is a continuous version of Embedded Voting for a large ensemble of estimators. Suppose the ensemble has many members but only \(\tau\) distinct complete types—each type captures an estimator’s score profile and its correlation embedding—and type \(t\) has population mass \(\mu_t\). For a finite candidate set, give each type a nonnegative score vector \(x_t\) and a unit embedding \(e_t\). With \(q\) retained dimensions, define
\[
G_j=\sum_t \mu_t x_t(c_j)e_te_t^\top
\]
for each candidate \(c_j\). The Continuous EV Winner problem is to return a candidate maximizing \(\prod_{\ell=1}^{q}\lambda_\ell(G_j)\), where the eigenvalues are ordered from largest to smallest. This is the mass-weighted counterpart of the paper’s product of squared singular values.
A plausible regime is a very large panel of similar forecasting systems or expert estimators, built from a small number of recurring methods or shared-information patterns. The number of estimators could greatly exceed the number of types, so mass captures their proportions. With finite rational support, I would expect approximate computation to be tractable when the winning scores are separated by a sufficient gap: the calculation reduces to \(m\times m\) matrices. That is a proposed follow-up, not a result of this paper.
The weak point is substantial: the paper studies selection quality under a noise model, not the computational complexity of finding a winner. Its embeddings are inferred from the observed candidate scores, and the experiments do not establish that few complete types adequately represent a large real ensemble. This makes the continuous EV question plausible, but it cannot supply the required hardness or algorithmic anchor.
There are no eligible in-paper anchors. Theorem 4.1 is cited from elsewhere, and this paper’s own contribution is a rule evaluated experimentally—not a named complexity result. Its results concern expected choice quality under a noise model. Turning them into a ChoCo question about worst-case algorithms would change the target, rather than mirror a result the paper establishes.
The proponent’s Continuous EV Winner makes that change concrete. Its matrix formula matches EV exactly if each type has the same realized score profile and embedding: the individual contributions \(s_i(c_j)e_i e_i^\top\) can then be grouped by type mass. But if types instead mean recurring estimator families, members have different realized scores. Replacing them with a type-level score requires a new limit model and assumptions about how scores and correlations are generated; it is no longer the paper’s finite-instance EV calculation. Encoding each realized score profile as part of the type preserves the calculation, but generally leaves nearly every agent as its own type. Either way, the proposed problem is new work, not a mirror of a named result.
A large panel built from a few recurring estimator families is plausible—the paper’s clone example itself suggests such a regime. That weakens any claim that a continuous population is meaningless here. The stronger negative is narrower: a worthwhile project on that regime would naturally study statistical estimation and performance guarantees for correlated ensembles. The paper offers no computational-complexity result to continue, and the proponent’s winner-computation problem does not supply one by mirroring an existing result. So I can defend “no worthwhile *anchored ChoCo mirror*,” but not the universal claim that no worthwhile continuous research question could arise from this 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.