| paper | d4-conf-wine-2025-0018 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | yes |
Theorem 3.1 characterizes strategic utility ratios; it does not assert that an object can be computed or that computing it is hard. The proposed \(\alpha\)-mass threshold problem is a coherent extension, but it cannot supply the missing computational anchor.
fails bit a — no named computational result to mirror
The proposed question concerns only the MNW collusion guarantees in Theorem 3.1; it leaves the PS and RR results and their outcome-level equivalence aside.
The strongest plausible population mirror is the collusion guarantee for MNW. Its anchor is Theorem 3.1, proved in this paper: with divisible goods and additive valuations, for every coalition-size bound \(c\ge1\), MNW has \(\mathrm{SGIR}(c)=c+1\) and \(\mathrm{GIR}(c)=2\). That is a precise result about strategic manipulation, but it does not meet your required anchor rule: it characterizes incentive ratios, rather than classifying a problem as polynomial-time, hard, FPT, or similar. The paper has no named result of that required computational-complexity kind.
A concrete extension would be \(\alpha\)-Mass MNW Collusion. An instance has \(m\) divisible goods and a finite set \(\mathcal T\) of complete valuation types, each with rational additive values \(v_t\) and rational mass \(\mu_t\), where \(\sum_t\mu_t=1\). Give each good one aggregate unit of supply; this is a per-capita scaling of the finite model, which leaves utility ratios unchanged. For a reported mass distribution \(\nu\), MNW chooses per-unit bundles \(x_r\) maximizing
\[
\sum_{r\in\mathcal T}\nu_r\log(v_r\cdot x_r)
\quad\text{subject to}\quad
\sum_{r\in\mathcal T}\nu_r x_{r,g}=1
\quad\text{for each good }g.
\]
A coalition is a nonzero mass flow \(z_{t,r}\): mass of true type \(t\) reports type \(r\), with \(\sum_r z_{t,r}\le\mu_t\) and \(\sum_{t,r}z_{t,r}\le\alpha\). The resulting reported mass is
\[
\nu_r=\mu_r-\sum_q z_{r,q}+\sum_t z_{t,r}.
\]
Each deviating unit of true type \(t\) reporting \(r\) gets utility \(u_{t,r}=v_t\cdot x_r(\nu)\); its truthful baseline is \(u_t=v_t\cdot x_t(\mu)\).
The threshold decision question is whether there is such a flow for which (i) every positive-flow group weakly improves and some group has \(u_{t,r}>R u_t\) (the SGIR version), or (ii) every positive-flow group has \(u_{t,r}>R u_t\) (the GIR version). The universal envelope over finite-support instances is the continuous analogue of the paper’s mechanism-wide ratios. A plausible regime is a large population competing for standardized resources, with many agents sharing a modest number of complete valuation types. That is a credible high-multiplicity setting; the continuum lies in the population, while the divisible allocation is a co-modification of the model.
The weakest point is also decisive for grading: Theorem 3.1 is not a computational-complexity anchor, and its finite-\(c\) formula does not directly give the answer for a coalition of mass \(\alpha\). It motivates the question, but it does not classify it. Natural follow-ups are whether continuous MNW has unbounded SGIR for every \(\alpha>0\), whether its GIR remains \(2\), and the complexity of the threshold problem in \(m,\tau,L\). I would leave its A/B/C classification open on the evidence in this paper.
Theorem 3.1 gives exact worst-case utility ratios for MNW as functions of coalition size. It contains no computational problem, input encoding, algorithm, or complexity bound. That is the decisive mismatch with ChoCo’s target: the theorem is a game-theoretic guarantee, so its population version would continue an analytic line of work rather than mirror a computational result.
The proposed \(\alpha\)-Mass MNW Collusion problem is coherent, and its population setting is plausible. It also introduces a new question. The theorem studies coalitions of at most fixed size \(c\); a coalition of mass \(\alpha>0\) corresponds, in a finite population of size \(n\), to \(c\) growing with \(n\). The theorem’s \(c+1\) bound then grows with the population. An \(\alpha\)-indexed guarantee would need a new scaling and a new analysis; it is not the continuous counterpart of the stated fixed-\(c\) result.
The threshold problem adds further choices: which reports are allowed, how reports are encoded, and how the mechanism selects among tied MNW allocations. Restricting reports to the existing finite type set gives a well-defined but narrower manipulation problem. Allowing arbitrary reports is also possible, but requires specifying a report domain and its encoding. Either could support a new complexity study; neither is supplied by Theorem 3.1, and neither turns its ratio characterization into a complexity result.
So I would reject this as an anchor for ChoCo, while granting that the mass model could motivate worthwhile work in its own right. The negative case is strongest against the paper as a source of computational mirrors; it cannot honestly establish that no valuable continuous study of collusion in fair division exists.
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.