| paper | d4-conf-aaai-1991-0064 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | yes |
The proposed source-population problem is coherent, and its agenda objective preserves the paper’s best-agenda computation. But the paper has no numbered theorem, lemma, corollary, or proposition asserting a computational result: its hardness equivalence appears in prose, while its approximation evidence is empirical. Under the required named-result gate, no anchor qualifies.
fails bit a — no named computational result to mirror
The only exact-hardness claim is unnumbered; the proponent explicitly acknowledges that this prevents it from serving as a compliant anchor.
fatal: True
The proposed mirror covers the paper’s prose best-agenda optimization claim. It does not provide a computational anchor for the empirical approximation results or the replanning discussion.
A credible mirror is Continuous Best-Agenda. Take a fixed rule set \(R\), and a large population of control sources. A type \(t\) records a source’s pairwise ordering preferences and their strengths; its mass \(\mu_t\) is the fraction of sources of that type. Choose an agenda \(\pi\), a permutation of \(R\), minimizing the total mass of contradicted evidence:
\[
\sum_t \mu_t\sum_{(a,b)}-\log(1-w_t(a,b))\,
\mathbf{1}[\pi\text{ places }b\text{ before }a],
\]
where \(w_t(a,b)\) is type \(t\)’s support for placing \(a\) before \(b\). A plausible regime is a large fleet of systems using a shared rule set, with many repeated control-source templates: the number of sources is large while the number of distinct preference types is bounded by the available templates. This preserves the paper’s agenda choice and its objective; it makes the source population continuous.
The expected result is hardness transfers. For a weighted directed graph on \(R\), make each arc \((a,b)\) a type with support \(1/2\) for \(a\) before \(b\), and mass proportional to that arc’s weight. Minimizing contradicted mass then gives exactly a minimum-weight feedback edge set. The hardness is driven by the number of agenda items, so this mirror does not make the optimization tractable. It covers the paper’s best-agenda problem, not its replanning or heuristic results.
There is a formal obstacle to making this a compliant anchored case: the paper contains no numbered Theorem, Lemma, Corollary, or Proposition asserting a computational result. Its “Optimization” section states in prose that maximizing plausibility is equivalent to weighted feedback edge set and says that problem is NP-complete, citing prior work. The approximation algorithm is presented with empirical evidence, not a numbered complexity result. So Continuous Best-Agenda is a plausible mirror of the paper’s central computation, but I cannot honestly supply the numbered anchor your criteria require.
The only plausible anchor is the paper’s best-agenda optimization: maximizing plausibility is equivalent to minimizing the weight of contradicted preferences, which the paper identifies with weighted feedback edge set. But the paper presents this in prose and cites the NP-completeness result; it does not state a numbered theorem, lemma, corollary, or proposition asserting a computational result. Its approximation result is empirical, and its replanning claims are not complexity results. Under the requested named-result criterion, the proponent has not supplied a qualifying anchor.
That is the strongest case against, but it depends on applying that criterion strictly. The proposed mirror itself is hard to dismiss. A population of control sources sharing a rule set and drawn from repeated preference templates is plausible; its type masses induce the weighted preference graph, and minimizing contradicted mass gives a well-posed continuous optimization question. The fact that its hardness may transfer, or that the continuous input can be summarized by edge weights, does not defeat the mirror under ChoCo’s stated standards.
So I cannot honestly make a strong universal case that no worthwhile continuous mirror exists. If the prose optimization result counts as an anchor despite not being a named result, this mirror survives the objections available 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.