| paper | d4-conf-ijcai-2007-0237 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | no |
Proposition 1 is a computational correctness guarantee for a named protocol, so bit (a) is met. But the only concrete continuous question is a planner’s full-information assignment problem, while Proposition 1 concerns bilateral strategic agents with hidden utilities, cooperative individual rationality, and sequential offers. The paper’s possible multilateral extension does not supply a continuous mirror of this anchored result.
fails bit b-changes — the continuum limit changes the question
The proponent acknowledges that the planner loses PONOMIR’s protocol and information constraints, but does not answer that the paper explicitly distinguishes its contribution from complete-information allocation work and centers private preferences and negotiation.
fatal: True
The proposed Type-Mass Pareto Agreement covers only a planner’s assignment of integral bundles to type mass. It does not cover Proposition 1’s bilateral negotiation protocol; the paper’s unnumbered [?2004l\(\mathcal{O}(2^H)\)[?2004l search bound and experimental search reductions supply no additional anchor.
The strongest positive case is for a narrow mirror of the paper’s Pareto-efficiency objective. The nearest named anchor is Proposition 1, proved here: under the paper’s cooperative-individual-rationality assumptions, PONOMIR produces a Pareto-optimal agreement. Those assumptions mean an agent avoids risky actions that could leave it worse off than the initial allocation, and, when two agreements give it equal utility, it accepts either one regardless of the other agent’s utility.
A plausible high-multiplicity setting is a large public programme assigning bundles of indivisible services or resources to applicants. Let each applicant’s complete type \(t\) specify a utility table \(u_t(S)\) for every bundle \(S\subseteq[H]\), plus a reservation utility \(r_t\). The instance gives rational type masses \(\mu_t\), with \(\sum_t\mu_t=1\), and per-capita supplies \(b_h\) of each resource \(h\). Many applicants share each type, so the number of applicants is large while the support size \(\tau\) is modest.
Call the resulting problem Type-Mass Pareto Agreement. For each type \(t\) and integral bundle \(S\), choose a mass \(x_{t,S}\ge 0\) of applicants who receive \(S\). Require \(\sum_S x_{t,S}=\mu_t\), resource use \(\sum_{t,S:h\in S}x_{t,S}\le b_h\), and \(x_{t,S}=0\) whenever \(u_t(S)<r_t\). The utility of type \(t\) is its average assigned utility,
\[
v_t(x)=\frac{1}{\mu_t}\sum_S x_{t,S}u_t(S).
\]
Given a feasible initial allocation, the task is to return an exact rational assignment \(x\) that is individually rational relative to that allocation and Pareto efficient over the type-average utilities: no feasible assignment improves every \(v_t\) and strictly improves at least one. Each person still receives an integral bundle; the fractions describe how mass of a type is distributed across bundles.
For explicit utility tables, this problem has a direct LP formulation over \(\tau 2^H\) type-bundle variables. Lexicographically maximizing the minimum type utility and then the sum of type utilities yields a Pareto-efficient solution. I would therefore expect this mirror to be tractable in \(\operatorname{poly}(\tau 2^H,L)\) time, where \(L\) is the input bit length. Compact utility representations would raise a separate pricing or separation question.
The fit to the paper is limited. Proposition 1 concerns two negotiating agents with incomplete information about each other’s preferences and a protocol designed to limit preference revelation. The proposed problem gives a planner the type utility tables and asks it to allocate resources across a population; it does not preserve PONOMIR’s protocol or information constraints. Also, Proposition 1 is a correctness guarantee, not a complexity-class result. The paper’s \(O(2^H)\) worst-case search claim and experimental reductions are not numbered complexity results. So this is a credible continuous-population analogue of the underlying Pareto-efficient resource-allocation goal, but it does not furnish an anchor-compliant complexity case for the paper’s main contribution.
Proposition 1 is the only named anchor, and it guarantees a property of PONOMIR’s output: two strategic agents reach a Pareto-optimal agreement under the paper’s cooperative-individual-rationality assumptions. It does not classify the complexity of a computational problem. The paper reports an \(\mathcal{O}(2^H)\) worst-case search bound and experimental reductions in search, but neither supplies a named complexity result for the proposed mirror to continue.
The Type-Mass Pareto Agreement problem replaces PONOMIR’s two private, strategic decision makers with a planner who knows everyone’s utility tables and chooses a global allocation. It drops the protocol’s distributed tree construction, sequential offers, and preference-revelation constraints. The paper distinguishes its contribution from complete-information resource-allocation work on just this ground. A planner’s type-mass allocation may be worthwhile in its own right, but it does not mirror Proposition 1. Its Pareto criterion also compares type averages: that can improve while some members of a type lose. A stronger formulation could track individuals’ starting bundles and mass transfers to preserve individual Pareto comparisons; that would repair this particular mismatch, while leaving the planner and full-information changes intact.
The best attempt to preserve the paper’s setting is to make each negotiating side a large constituency with finitely many complete preference types. Then the population masses describe constituents, while the two principals still make the offers and decisions. That makes high multiplicity plausible, but it changes the principals’ aggregate utility functions rather than making the negotiating agents themselves a continuous society. If instead each individual becomes a negotiating agent, the model needs new rules for who proposes, how acceptance works, and how private preferences are handled. The paper’s stated guarantee does not determine those rules or their computational problem.
That leaves a real limit to the negative case: the paper discusses large, diverse agent populations and says PONOMIR can extend to multilateral negotiation. A high-multiplicity version of that extension could be sensible and worth studying. But the paper gives no named result for it, and its multilateral claim is not developed into the anchor the proponent needs. I can reject the proposed mirror as a mirror of the paper’s named result; I cannot honestly rule out every worthwhile continuous mirror.
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.