| paper | d4-journals-jair-2018-0042 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | unclear |
Theorem 5 and Theorem 9 are constructive correctness results, so they have computational character even without runtime bounds. But the proponent gives no precise population-distribution problem for either result: under the paper’s set-based semantics, duplicate reports add no mass-sensitive input, and making prevalence affect merging would require a new aggregation model.
fails bit b-none — no population with multiplicity to relax
No result is covered by a surviving mirror. Theorem 5 and Theorem 9 give correctness results for finite source sets; Section 6.3’s weighted-distance examples do not supply an anchored computational population result.
I can’t make a compliant anchor based case for a continuous mirror. The paper gives algorithms, but no numbered result classifies a problem as polynomial time, hard, FPT, or otherwise.
The closest results are Theorem 5, proved here, which establishes that Algorithm 1 outputs a weakly stable two class partition and a maxcon that induces it; and Theorem 9, proved here, which establishes that Algorithm 2 outputs a weakly stable and a strongly stable partition. Neither gives a runtime bound. Algorithm 1 uses propositional consistency checks, and Algorithm 2 searches among maxcons, so their correctness does not establish tractability.
A continuum of users sharing complete belief report types is imaginable, but turning their population masses into a merging rule would require choices about how multiplicity affects maxcons and reliability. The paper’s results do not validate those choices or provide a computational result to mirror.
The strongest case against a mirror of the cited results is that neither anchor states a computational complexity result. Theorem 5 proves that Algorithm 1 outputs a weakly stable bipartition and an inducing maxcon; Theorem 9 proves the corresponding correctness claim for Algorithm 2. Neither gives a running-time bound, a hardness result, or an approximation guarantee. Section 8.6 explicitly leaves the complexity of related tasks open. A continuous version of those algorithms could be a new question, but it would not mirror a complexity conclusion established by this paper.
There is also a specific obstacle to continuizing those results. A complete type would be a source’s entire set of formulas. The maxcons and induced partitions depend on unions of formulas and consistency. Duplicating a source of the same type adds no new formulas, and every copy is classified the same way. So changing the masses while keeping the support fixed does not change the merge or its stability. For Theorems 5 and 9, a population distribution collapses to its support; it contributes no population variable for continuous optimization to act on.
The best repair is to use the paper’s weighted-distance framework: let \(\mu_t\) be the mass of source type \(t\), and evaluate interpretations using an aggregate distance such as \(\sum_t \mu_t d_t(I)\). That makes repetition affect the result and could suit a crowd of independent reporters with recurring report types. The paper’s Section 6.3 gives this direction some footing, so I cannot rule out a worthwhile continuous problem that its authors would recognize. But the section offers examples of stable weights, not a named computational result; making mass count as influence also changes the aggregation model.
So the negative case is strong against Theorems 5 and 9 as anchors: their exact population versions are support-only, and their algorithmic results have no complexity classification to continuize. The broader claim that no scenario could yield a worthwhile mirror is weaker. A mass-weighted version of the paper’s own framework is plausible, though it would be a new computational question rather than a mirror of a result proved here.
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.