| paper | d4-conf-atal-2023-0275 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | unclear |
Neither case identifies a numbered theorem, lemma, corollary, or proposition with a computational claim: Equation (1) defines an objective, Algorithm 1 gives a heuristic, and the experiments state no guarantee. The proposed type-mass consultation problem is a prospective reformulation, not a continuous analogue anchored in a computational result of this paper.
fails bit a — no named computational result to mirror
No named computational result is available to anchor a mirror; the paper's selection heuristic and empirical comparisons do not meet the stated anchor criterion.
MADDM contains no admissible named computational anchor: no numbered theorem, lemma, corollary, or proposition establishes a complexity or tractability result. Equation (1) defines the utility-maximizing selection function, Algorithm 1 gives a selection heuristic, and Section 4 reports empirical results on simulated finite advisor sets. None asserts that the underlying problem is in P, hard, FPT, or another complexity class.
So there are no anchor-specific continuous problems to state, and I can’t responsibly assign a Class A/B/C expectation. The most plausible future mirror would treat a large pool of interchangeable advisors—such as instances of the same model with the same price and trust evidence—as types with masses \(\mu_t\), then optimize how much of each type to consult across a stream of decisions. That connects naturally to MADDM’s advisor-selection objective, but it is a prospective research question, not a result this paper establishes. The main weakness is that MADDM’s trust estimates are tracked for individual advisors; grouping them into types could lose distinctions as their histories diverge.
MADDM has no named computational result to mirror. Equation (1) defines an optimization objective; Algorithm 1 gives a Thompson-sampling heuristic; Section 4 reports experiments. The paper proves no complexity or tractability claim. Treating Equation (1) as an anchor would require first specifying an exact computational problem the paper does not analyze.
The strongest candidate is a large crowdsourcing pool, with workers grouped by price and reliability characteristics, where a decision-maker chooses how much of each group to consult across repeated tasks. That is a plausible setting, and the authors might recognize it. But it is a new model: MADDM’s key state is each advisor’s trust estimate, updated from that advisor’s history using a truth estimate that depends on the other selected advisors. Pooling workers either changes that learning rule or requires types that track evolving histories. Replacing discrete consultations with queried mass also changes the evidence and exploration process. These are substantial changes to the object being optimized, even if they produce a useful research question.
That is the strongest case against calling this a continuous mirror of MADDM: the paper gives no computational theorem to carry across, and its most natural population-scale variant would need a new formalization of the dynamic learning problem. But it does not support the universal claim that no worthwhile mirror exists in any scenario. Repeated high-volume annotation with interchangeable worker groups is credible, and a continuous consultation model could be worth studying on its own. The negative case is strong against an anchor-based mirror of this paper, but weak against that broader possibility.
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.