| paper | d4-conf-ecai-2025-0675 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | medium |
| authors would recognise it | yes |
P→P discrete result → its continuous mirror
Answer: Polynomial-time solvable in \(m,\tau,L,p\), uniformly over all four semantics. · discrete side: polynomial · adjudication: proved
full proof & adversarial review →
Lemma 1
statement extracted from the paper’s text layer
Given an acyclic debate graph, a semantics \(\sigma\), a finite set of complete response types \(T\), a rational distribution \(\mu\) over \(T\), a population scale \(N\), and an IDM parameter \(\beta\), compute the final acceptability interval for every claim. Each type records, for every claim, abstention or a score in \(\{0,1,2,3,4\}\); use vote masses \(k_{a,l}=N\sum_{t:t(a)=l}\mu_t\) to form the paper's normalized IDM initial intervals, then propagate their bounds using \(\sigma\).
A large consultation with a fixed agenda has a distribution \(\mu\) over complete response types, each carrying mass. Per-claim vote masses are the marginals of \(\mu\); the IDM uses those masses with \(\beta\) as pseudo-mass, and the paper's semantics propagate the resulting intervals through the debate graph. The full joint type distribution may be compressed to the per-claim vote histograms for this computation.
With fixed IDM parameter \(\beta\), positive per-capita vote participation makes the intervals narrow as population size grows; keeping finite uncertainty may require vanishing participation or a population-scaled \(\beta\). This limits which high-multiplicity regimes retain the paper's uncertainty question.
fatal: False
The mirror covers Lemma 1's endpoint computation for IDM-initialized intervals over an acyclic debate graph. It does not establish that Proposition 4's uncertainty amplification, or the paper's other general interval properties, persist when initial intervals are restricted to those generated from population vote masses.
The strongest positive case is narrow: this paper has a plausible population mirror, but no eligible named computational anchor under your rule. Propositions 1–6 and Lemma 1 establish semantic properties and interval calculations; none classifies a problem as in P, NP-hard, or similar. The authors say in §3.1 that interval computation is polynomial on acyclic graphs, but that claim is prose, not a numbered complexity result. So I can’t honestly present an anchor-backed case.
The nearest named results suggest a concrete follow-on. Lemma 1, proved here, says that when a semantics satisfies bi-variate equivalence, proportionality, and reinforcement, the lower final bound is obtained using the lowest initial weight, strongest attackers, and weakest supporters; the upper bound uses the highest initial weight, weakest attackers, and strongest supporters. Proposition 4, also proved here, says that QuAD, DF-QuAD, Exb, and QuEM all fail uncertainty reduction via semantics computation: final interval width can exceed initial width, even in frameworks containing only attack relations or only support relations. Proposition 4 is the closer substantive anchor, but it is a property result, not a computational one.
A plausible lead candidate, not an eligible anchor, is *Mass-IDM Interval Propagation*. Consider a large public consultation on a fixed, moderate agenda of claims. A voter type is their complete response vector across the agenda, with each entry either abstention or a score in \(\{0,1,2,3,4\}\). The society is a distribution \(\mu\) over those types; for a consultation of size \(N\), type \(t\) represents mass \(N\mu_t\). A regime with \(N\gg\tau\) is credible where standardized ballot patterns recur, though it is a narrower scenario than unrestricted online debate.
Given an acyclic debate graph, a semantics \(\sigma\), rational \(\mu\), \(N\), and IDM parameter \(\beta\), compute each claim’s vote masses \(k_{a,l}=N\sum_{t:t(a)=l}\mu_t\). Use those masses in the paper’s IDM expected-score bounds, normalized to \([0,1]\), as initial intervals; then propagate them under \(\sigma\). A solution is the final interval for every claim. One associated decision question is whether some claim’s final interval is wider than its initial interval. The interpretation of \(\beta\) as pseudo-mass is a necessary co-modification: it preserves the paper’s uncertainty model when votes are represented as mass.
The authors would likely recognize this as their problem in a high-multiplicity consultation regime: the response categories, sparse participation, debate graph, and gradual semantics all remain. The expected direction is tractable, following the paper’s acyclic evaluation procedure and Lemma 1’s endpoint reduction. Whether Proposition 4’s uncertainty amplification persists when the intervals must come specifically from IDM-generated population masses is a further question; Proposition 4 alone does not establish it. Other useful questions are how amplification varies with \(N\) and \(\beta\), and whether one can bound it.
The soft spot is substantial. Full response vectors over a large debate can have nearly one distinct type per person, defeating the small-\(\tau\) regime. The fixed-agenda consultation makes \(N\gg\tau\) plausible, but it is narrower than the paper’s Kialo setting. Also, the paper already gives efficient acyclic evaluation, so this mirror is chiefly a high-multiplicity formulation of its uncertainty analysis. It does not expose a discrete hardness result whose fate under continuization could be studied.
The negative case is unusually strong at the programme’s entry gate: this paper has no named computational result to mirror. Lemma 1 is an endpoint-reduction result, and Proposition 4 is a property of four semantics. Neither formulates or classifies a computational problem. The paper does say that interval propagation on acyclic graphs is polynomial time, but that is not a numbered complexity result; moreover, it already gives the evaluation procedure. So the proponent’s own proposed anchor is a follow-on question, not a mirror of an eligible result.
I would not reject the proposed consultation scenario just because it uses repeated ballot patterns. That is a sensible high-multiplicity regime. The difficulty is that, for the paper’s task, a distribution over complete response types contributes only per-argument vote counts. Once those marginals are known, the correlations between voters’ responses across claims play no role: the IDM produces each claim’s initial interval from its own vote histogram, and the semantics propagate those intervals through the graph. The proposed question—compute final intervals and check whether any widened—is therefore already handled by the paper’s acyclic propagation procedure. The mass representation changes how the histograms are supplied, but creates no new pricing, optimization, or other computational subproblem.
The stronger version, making the population distribution itself the unknown and asking how sparse observations constrain possible final intervals, could require substantial optimization over joint response types. But that changes the object being studied. The paper applies IDM to observed votes on each argument; it does not infer a population distribution over complete ballots. That extension might be worthwhile research, but it is not a continuous mirror of Lemma 1 or Proposition 4.
There is also a scaling tension in preserving the paper’s uncertainty model. With fixed IDM parameter \(\beta\), a claim’s normalized initial interval has width \(\beta/(K_a+\beta)\), where \(K_a\) is its number of observed votes. If vote participation for that claim stays positive as population size \(N\) grows, then \(K_a\) grows with \(N\) and the interval narrows to zero. If \(K_a\) stays bounded, its vote fraction vanishes; the limiting population proportions alone no longer retain the finite vote counts needed by the IDM. Keeping those counts and \(N\) as extra inputs preserves the computation, but leaves the population distribution with no new computational role.
That scaling point is not a proof that no conceivable population model for deliberation is useful. The weaker and more defensible conclusion is that the case offered here does not identify a worthwhile computational mirror: its named results are semantic properties, its candidate computation is already covered by the paper’s algorithm, and making population uncertainty operational requires adding a different inference problem.
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.