| paper | d4-conf-aaai-2026-1406 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | yes |
The degree-mixture prediction is a faithful way to restate Theorem 3's asymptotic quantities. But evaluating those explicit formulas does not turn the paper's analytic phase-transition result into a computational result, so no anchor passes the computational gate.
fails bit a — no named computational result to mirror
The proposed prediction covers Theorem 3's threshold and limiting red-density formulas over degree types; it leaves Proposition 1's duality and Conjecture 5's consensus-time question aside.
There is no named computational-complexity result in this paper to anchor a ChoCo mirror: no theorem or proposition classifies an underlying problem as tractable, hard, or parameterized. The strongest positive case is therefore conditional. The paper does give a named asymptotic result that can be restated with a continuous population as input, but that is an analytic prediction problem, not the computational gap ChoCo targets.
A plausible regime is a large online platform with many users but relatively few degree classes. Let a type be a user’s in-degree and out-degree, \((a,b)\), with the paper’s update rule and bias \(p\) shared across types. Let \(\mu_{a,b}\) be the fraction of users of that type. A reasonable instance regime has \(n\gg\tau\), where \(\tau\) is the number of supported degree pairs, and the degree bounds and moment conditions of the paper hold. Mass is the fraction of users in each degree class; the network is sampled by the directed configuration model. This is plausible for a platform with many users in recurring degree strata, though it is a stylized model of who interacts with whom.
One precise candidate is Continuous Degree-Mixture Tipping Prediction. Its input is a finite-support rational distribution \(\mu\) on degree pairs \((a,b)\), with \(b\ge2\), balanced mean degrees \(\sum_{a,b}\mu_{a,b}a=\sum_{a,b}\mu_{a,b}b=\bar d\), and a rational bias \(p\). Its output is whether the system is above or below the transition and the limiting red fraction from an initially all-red society. Define
\[
\varrho(\mu)=\frac{1}{\bar d}\sum_{a,b}\mu_{a,b}\frac{a}{b},
\qquad
\lambda(\mu)=\sum_{a,b}\mu_{a,b}\frac{1}{b}.
\]
The answer is the transition value
\[
p_c(\varrho)=\frac{\sqrt{1-\varrho}-(1-\varrho)}{\varrho}
\]
and limiting red mass \(0\) for \(p\ge p_c\), or
\[
q_\star(p,\varrho,\lambda)=
\left(1-\frac{p^2}{(1-p)^2(1-\varrho)}\right)
\left(1-\frac{p^2(\lambda-\varrho)}{1-\varrho}\right)
\]
for \(p<p_c\). A solution is this classification and value. It is a prediction task, not an optimization problem: \(p\) is an input, not a chosen intervention. The explicit formulas make this candidate tractable by direct computation, but the paper does not state or analyze it as a complexity problem.
The relevant named result is Theorem 3, proved here. In full, for degree sequences satisfying the paper’s conditions (equal total in- and out-degree, minimum out-degree at least \(2\), bounded maximum out-degree, bounded \((2+\delta)\)-moment of in-degrees, and convergence of \(\varrho_n\to\varrho\) and \(\lambda_n\to\lambda\)), take any slowly diverging \(t_n\). Under the joint law of the directed configuration model and the opinion dynamics started from all red, if \(p\ge p_c(\varrho)\), then for every \(\varepsilon>0\) the probability that the red fraction at \(t_n\) is below \(\varepsilon\) tends to \(1\). If \(p<p_c(\varrho)\), then for every \(\varepsilon>0\) the probability that the red fraction at \(t_n\) differs from \(q_\star(p,\varrho,\lambda)\) by less than \(\varepsilon\) tends to \(1\). The theorem supplies the formulas above; it does not assert a complexity classification. Proposition 1’s duality with COBRAD is useful analytic machinery, but it also gives no computational result, so I would not count it as another anchor.
The positive case is that a degree-type distribution is a natural population description, and Theorem 3 shows that this aggregate description suffices for a meaningful question about adoption at large scale. The weakness is decisive for ChoCo: the paper already studies the \(n\to\infty\) behavior and proves the aggregate limit. Restating its degree sequences as a measure does not uncover a new computational landscape. The authors would likely recognize the prediction question, but as their existing asymptotic analysis, not as a new computational-social-choice problem.
Theorem 3 is the only plausible anchor, and it does not state a computational problem. It proves an asymptotic law for the opinion process on random directed configuration graphs: at slowly diverging times, the red fraction concentrates at a value determined by degree statistics. Proposition 1 supplies analytic machinery for that result, and Conjecture 5 concerns consensus times. None asks for an algorithm, classifies complexity, or studies an intervention problem.
The proponent’s degree-mixture prediction is a sensible continuous *description* of the theorem’s input. But its output is the theorem’s asymptotic prediction, not a computational question the paper poses. Turning it into “compute the threshold and limiting fraction from the degree masses” would wrap the stated formulas in an input-output format; it would not expose a computational problem native to the paper. The theorem’s large-graph limit is already analytic. That does not count as prior computational work under ChoCo’s rules; it means this anchor supplies no computational result for ChoCo to mirror.
The strongest extension would make the disruptive bias a policy choice: choose the least uniform bias needed to reach a target adoption level, or allocate intervention across degree types. The uniform version adds only a scalar choice to the paper’s explicit prediction. The type-targeted version could lead to an interesting model, but it changes the update rule and its dual process. The paper does not establish that problem or its computational structure, so it is a possible new research direction rather than a mirror of Theorem 3.
That leaves the negative case strong on the paper’s actual results and weaker on the universal claim. The paper’s references to marketing and intervention make type-targeted bias a plausible extension an author might recognize. I cannot rule out that it becomes a worthwhile computational-social-choice question after substantial modeling work. It does not, however, rescue the sole anchor raised here: the paper itself gives no named computational result to continuize.
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.