| paper | d4-conf-ijcai-2017-0575 |
| 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 has no named computational result to anchor a mirror: Equation (4) is an objective used by the agent, while the reported performance comparisons are empirical. The proposed type-mass LP replaces predicted individual acceptance and coalition unanimity with deterministic reservation thresholds, making it a related follow-up rather than a mirror of a result this paper proves.
fails bit a — no named computational result to mirror
No named computational result is mirrored; the proposed LP leaves out learned response prediction, the EDP model, and the experimental findings.
This paper has no numbered Theorem, Lemma, Corollary, or Proposition asserting a computational result. It gives definitions, a predictive model, an optimization objective, and experimental findings, but no qualifying anchor. So there are zero anchor-backed mirror questions to state.
The closest foothold is Equation (4): the EDP agent chooses a payoff division \(\vec{x}\) to maximize \(x_p\prod_{i\in\operatorname{supp}(\vec{x}),\,i\ne p}\Pr(\mathrm{Acc}_i\mid\vec{x},p)\). The paper approximates this by enumerating payoff divisions in five-unit increments in its five-agent setup. It reports that the agent earned an average proposer share of \(43.78\), compared with \(27.26\) for human proposers. Those are empirical results, not numbered computational theorems.
A narrow, unanchored continuous analogue would model a large bargaining constituency with \(N\gg\tau\) agents in \(\tau\) repeated types. A type \(t\) records a member’s voting weight \(w_t\) and a common minimum acceptable offer \(\rho_t\); its rational mass is \(\mu_t\). Given a quota \(q\), reward \(r\), and a focal proposer whose contribution has vanishing normalized weight, choose winning coalition masses \(y_t\), total transfers \(b_t\), and proposer payoff \(z\), subject to \(0\le y_t\le\mu_t\), \(b_t\ge\rho_t y_t\), \(\sum_t w_ty_t\ge q\), and \(z+\sum_t b_t=r\). Maximize \(z\), returning the coalition masses and transfers. This is a linear program in \(\tau\) types, so this simplified problem is Class A.
A large membership association or cooperative with many interchangeable voters and a small number of bargaining types could make that population regime plausible. The model retains the quota, winning coalition, and revenue division. But it replaces the paper’s central feature—individual probabilistic acceptance and power-based prediction—with a common deterministic reservation offer for each type. That is a substantial change, and the original authors might regard it as a related coalition-purchasing problem rather than the continuous version of their EDP agent. Keeping independent acceptance probabilities also raises a limit problem: the chance that every member of a growing coalition accepts can collapse unless offers make each acceptance probability approach one. How to preserve that behavior, and how to extend the EDP index’s uniform choice over finitely many minimal winning coalitions, are open modeling questions.
So the positive case is limited: there is a plausible high-multiplicity question near the paper’s proposer-optimization objective, but no qualifying named result anchors it. The proposed LP covers only a simplified proposer-side coalition problem; it does not mirror the paper’s learning results or experimental claims.
The decisive negative is that this paper supplies no named computational result to mirror. Its numbered sections report definitions, a learned acceptance model, a finite-grid search in the five-agent experiment, and empirical comparisons. Equation (4) is the best foothold, but it states the agent’s optimization objective; the paper gives no complexity theorem for that optimization problem. The reported revenue difference, \(43.78\) versus \(27.26\), is an experimental finding, not a computational result. So the proponent has identified a possible follow-up, not a surviving anchor from the paper.
Their LP makes a coherent model when each type has a deterministic reservation offer. It also replaces the paper’s central behavioral feature: predicting whether particular responders accept particular offers. In the paper’s game, a proposed coalition succeeds only if every designated member accepts. If a positive mass of repeated types has independent acceptance probability \(a_t<1\), selecting \(Ny_t\) members of type \(t\) gives success probability \(\prod_t a_t^{Ny_t}\), which tends to zero as the population grows. Keeping the stochastic acceptance model therefore makes the coalition-success objective collapse unless offers drive each member’s acceptance probability to within \(O(1/N)\) of certainty. A model where enough accepted weight suffices avoids that collapse, but changes the paper’s unanimity condition and becomes a different bargaining question.
A stronger repair would represent stable individual differences as types, perhaps by weight and reservation value, and optimize over the masses of types willing to join. That is a sensible high-multiplicity weighted-voting problem, and it could be worth studying on its own. It no longer tests the paper’s EDP-based prediction of human responses. Extending EDP also needs a principled population limit: its definition averages uniformly over finite minimal winning coalitions, and a continuum of people has no uniform distribution over such coalitions without specifying how that limit is constructed.
So the negative case is strong for treating this paper as an anchored ChoCo candidate: it has no qualifying theorem, and its most promising mass formulation sheds the empirical mechanism that motivates the paper. I cannot honestly make the universal claim that no scenario yields a worthwhile continuous problem. A large population with typed reservation values could support one; the paper simply gives little basis for calling it a mirror of a result it proves.
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.