| paper | d4-journals-mss-2020-0061 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | unclear |
The paper’s numbered propositions characterize the indices and their zero cases; none states an algorithm, complexity bound, hardness result, or computational formulation used to compute. The evaluator-panel threshold is a plausible follow-on question, but the paper neither establishes it as a result nor analyzes its computation, so no named result meets bit (a).
fails bit a — no named computational result to mirror
The paper defines the Manhattan and Kendall inconsistency indices for individual incomplete pairwise-comparison matrices, gives their properties, and compares indices empirically; it contains no population or computational-complexity result.
The strongest plausible population setting is a large standardized evaluation panel. A type would be an evaluator’s complete response: one incomplete reciprocal pairwise comparison matrix over a fixed alternative set \(A\). The mass \(\mu_t\) would be the fraction of the panel with response type \(t\). A natural population question is: given finitely many such types, rational masses summing to one, and a threshold \(q\), is the panel’s mean Manhattan inconsistency at most \(q\), that is, does \(\sum_t \mu_t\,\mathrm{MII}(M_t)\le q\)? This keeps the paper’s index and makes the population itself continuous. A procurement review panel or a large crowdsourced evaluation could make repeated response types plausible.
But this is only a proposed extension, not a result the paper establishes. The paper contains no named computational result asserting a problem is in P, NP-hard, or belongs to another complexity class. Proposition 4.3 proves that \(\mathrm{MII}(M)=0\) exactly when \(M\) is consistent; Proposition 4.11 proves that consistency implies \(\mathrm{KII}(M)=0\), while Example 4.12 shows the converse fails. These are properties of the indices, not computational-complexity results. The Monte Carlo comparison is empirical and supplies no named complexity anchor.
So there is no qualifying anchor here, and no continuous problem can honestly be presented as the mirror of one of the paper’s named computational results. For the proposed panel question, I would expect the main computational challenge to remain the spanning-tree calculations for each type; averaging over population mass does not by itself resolve it. The weakest point is therefore fundamental: the population model is plausible as a follow-on application, but this paper studies inconsistency of matrices, not a computational problem about a population of evaluators.
The strongest negative is that this paper has no named computational result to continuize. Proposition 4.3 characterizes when MII is zero; Proposition 4.11 gives a one-way implication for KII, and Example 4.12 shows its converse can fail. None asserts a complexity bound or analyzes an algorithmic problem. The Monte Carlo section reports empirical comparisons, not a complexity result. Those references therefore do not provide computational anchors.
The evaluator-panel proposal is a plausible follow-on application, but its mean-MII threshold question is newly introduced rather than a mirror of one of those results. For a finite support of matrix types, the population score is \(\sum_t \mu_t\,\mathrm{MII}(M_t)\). The zero case follows immediately from Proposition 4.3 and nonnegative masses; other thresholds require scoring the matrices and taking a weighted sum. Any harder computation is the per-matrix index calculation, which this paper does not analyze as a complexity problem. Treating the Monte Carlo matrices in Section 5 as a population gives a similar extension: it replaces sample averages with expectations, but the objects being distributed are matrix instances, not agents in a society.
One could add intervention costs and ask how to change enough evaluators’ matrices to lower the mean inconsistency. That would create an optimization problem, but its costs, interventions, and objective are not supplied by the paper. It is a new model, not a continuous mirror of a result here.
The panel setting is the best case for a useful population application, so the universal claim that no scenario could ever be worthwhile is hard to prove. But on the programme’s stated test—finding a computational result in this paper whose continuous-population version is worth studying—the negative case is strong: none of the cited propositions, example, or simulations qualifies.
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.