| paper | d4-conf-ecai-2012-0105 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | unclear |
Corollary 1 gives a semantic characterization of \(Y\)-entailment, but asserts no algorithm, runtime bound, hardness, or computational formulation. The finite-support share is well-defined, and cited prior work [16] appears to make it polynomial by applying individual inference per type, but that is an external answer to an added analytics question, not a continuous mirror of a computational result established here.
fails bit a — no named computational result to mirror
The proposed share uses the per-type \(Y\)-entailment test characterized by Corollary 1; it leaves the paper's other semantic characterizations and establishes no population-level computation.
The best positive lead I can defend is a large recommender service with many users who share the same complete set of comparative preference statements. A type is one such statement set \(\Gamma_i\), together with the model parameters; its mass \(\mu_i\) is the fraction of users of that type. Given a common pair of outcomes \(\alpha,\beta\) and a valid family \(Y\), define the certified preference share problem: from the finite support \(\{(\Gamma_i,\mu_i)\}_{i=1}^{\tau}\), compute exactly
\[
M_{\alpha\succeq\beta}=\sum_{i:\,\Gamma_i\models_Y\alpha\ge\beta}\mu_i.
\]
The output is a rational mass. This keeps the paper’s individual inference question intact, then asks what fraction of a large, repetitive user population receives that inference. The cohort scenario is plausible given the paper’s recommender-system motivation, though the paper studies users individually.
The nearest named result is Corollary 1, proved here: \(\Gamma\models_Y\alpha\ge\beta\) if and only if every weak order satisfying \(\Gamma\) and overall importance with respect to \(Y\) also satisfies \(\alpha\ge\beta\). This precisely characterizes the inference used by the proposed population problem. But it does not assert a complexity class. The paper’s polynomial-time claim appears in the introduction as prior work cited to [16], rather than as a numbered complexity result proved here. So Corollary 1 is a semantic anchor, not an anchor that meets the task’s computational-result requirement.
If the cited polynomial inference algorithm applies to each explicitly listed \(\Gamma_i\), this aggregate problem is tractable by running it once per support type and summing the corresponding masses. Its running time would be polynomial in the total explicit input size, including \(\tau\) and the rational masses’ bit lengths. That is a plausible extension, not a result of this paper. It raises questions about compressed support and threshold queries such as whether \(M_{\alpha\succeq\beta}\ge q\).
The weak point is decisive under the stated anchor rule: the paper’s numbered results characterize preference semantics and importance; none asserts that an underlying problem is in P, NP-hard, or otherwise classifies its complexity. The proposed cohort question is therefore a modest, author-recognizable population extension, but not a fully anchor-compliant continuous mirror of a named computational-complexity result.
The decisive problem is that this paper has no numbered result classifying the complexity of a computational problem. Theorem 1 and Corollary 1 characterize which preference orders are generated by \(Y\)-cp-trees; Theorem 2 and Corollary 2 give another semantic characterization. The polynomial-time inference claim is attributed to prior work [16]. So Corollary 1 is a result one can anchor a question to, but it is not itself the computational result ChoCo is looking to continuize.
The recommender-service scenario is plausible, and the proposed certified preference share is well-defined. Users with the same complete statement set and parameters can be a type, and their mass can be meaningful. But the share simply aggregates individual entailment answers across types. The society distribution does not enter the entailment semantics or couple the users’ problems. That makes this an outer population-analytics task, not a continuous-population mirror of a computational result established here.
A richer version—say, choosing an intervention to increase the mass of users for whom a preference can be inferred—could introduce a substantive optimization problem. But the intervention, its costs, and its effect on users’ statement sets would all be new structure absent from the paper. It may be worth studying on its own; it cannot rescue an anchor to this paper’s results. Even if the cited polynomial inference algorithm makes the proposed aggregate easy to compute, ease is not the objection: the missing link is that the paper establishes no computational complexity result for the population question or its per-user kernel.
So the proponent identifies a sensible high-multiplicity extension, but not a qualifying mirror. This negative case is strongest on that scope boundary; it does not show that population-level preference analytics could never be worthwhile.
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.