| paper | d4-conf-atal-2021-0220 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | no |
The paper reports simulation correlations and an analytical identity, but no numbered theorem, lemma, corollary, or proposition asserting a computational result, so bit (a) fails. The proposed mass recurrence also averages away local poll classifications and shifts the task from comparing predictors on network simulations to computing outcomes in a dense mean-field model.
fails bit a — no named computational result to mirror
No named computational result is covered. The proposed recurrence would compute final red mass, but leaves out the paper’s comparison of predictors across randomized networks and behavioral runs.
There is no admissible anchor in this paper under your rule. It contains no numbered Theorem, Lemma, Corollary, or Proposition asserting a computational complexity result. Its findings are simulation results—such as the weak influence-gap correlation in Figure 4 and the stronger initial-majority correlation in Figure 6—and an analytical calculation, Equation (7), showing that influence gap is zero on certain equal-size clique graphs. Those are not named complexity results. The cited PSPACE-completeness result on opinion-diffusion convergence is from other work and concerns a different question.
A plausible continuous question, though, is not anchored to a qualifying result: consider a large discussion forum split into a small number of partisan communities, with users sharing their community, initial party, and behavioral strategy. Let each type \(t\) have rational mass \(\mu_t\), strategy parameters \(\theta_t\), and interaction rates \(W_{tu}\) with other types. Given the paper’s voting threshold \(V\) and horizon \(H\), track \(q_t^r\), the fraction of type \(t\) voting red at round \(r\). Each type’s poll is determined by the kernel-weighted red mass among its contacts; the paper’s win, lose, or deadlock classification and strategy probabilities then determine \(q_t^{r+1}\). The problem is to compute the final red mass \(\sum_t \mu_t q_t^H\) and decide whether it reaches \(V\).
The regime would be many users but relatively few community, party, and strategy types. In a high-degree, exchangeable-block regime, this looks like a plausible population version of the paper’s outcome-prediction question; the type-level recurrence suggests a tractable answer. The weak point is substantial: it averages away the local neighborhood variation and correlations that the paper studies. It is therefore a candidate mean-field mirror, not a basis for claiming that any named result of this paper has a continuous counterpart.
The strongest negative starts with a gap in the case for: this paper has no numbered theorem, lemma, or corollary about the complexity of a computational problem. Its own results are simulation findings about predictive correlations and an analytical identity, Equation (7), for equal representation on isolated cliques. The PSPACE-completeness result it cites is from other work and concerns convergence, not this paper’s prediction question. So there is no qualifying result here for a continuous mirror to inherit.
The proposed block-type recurrence is the best candidate anyway, but its type averages do not generally preserve the dynamics the paper studies. Each voter responds to the win, lose, or deadlock classification of their own neighborhood poll. Replacing that poll with its type-level expected share can change the classification, and averaging also loses correlations among voters who share neighbors. In sparse networks those local effects persist as the population grows; representing them faithfully requires neighborhood-level state, which can make the type space as large as the original network. In dense exchangeable blocks, polls can concentrate enough for the recurrence to work, but then the question becomes a mean-field analysis of a new limiting model. Computing final vote share also leaves out the paper’s central empirical task: comparing predictors across randomized networks and behavioral runs.
A richer version could keep a graphon or a distribution over local network structures and ask for algorithms to predict the final vote share. That might be worthwhile research, but its computational input and question would be newly supplied; they are not a mirror of a complexity result in this paper. The negative case is strongest as a verdict about this paper’s suitability for ChoCo. It is weaker as a universal claim: a large, exchangeable block-network setting is plausible, and a mean-field prediction problem there could be useful to the broader field.
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.