| paper | Constrained Fair and Efficient Allocations |
| authors | Benjamin Cookson, Soroush Ebadian, Nisarg Shah |
| venue | AAAI 2025 |
| filed under | fairalloc · indivisible |
| judged by | gpt-5.6-luna / xhigh (triple__luna__xhigh__c2r1) |
| judge confidence | high |
| authors would recognise it | yes |
Theorem 4 is a named result, but it asserts fairness and efficiency guarantees rather than computational complexity or an algorithm. The proposed continuum-MNW formulation is a sensible, author-recognizable high-multiplicity follow-up, but it is not a mirror of a computational result in this paper. Therefore bit (a) fails and the grade is red.
fails bit a — no named computational result to mirror
The proposed model covers the goods-with-copies MNW guarantee of Theorem 4, while leaving the other fairness, impossibility, equilibrium, and best-of-both-worlds results untouched.
The strict verdict is that this paper contains no valid computational anchor under the ChoCo rule. The numbered results are existence, fairness, efficiency, or impossibility statements:
Section 5 even says that the paper does not discuss computing MNW allocations and only conjectures that the problem is hard, citing Lee (2015). That cited hardness is not a numbered result of this paper. Thus there is no honest “one continuous problem per computational anchor” to present. A positive computational-mirror case would be a follow-up to this paper, not a mirror anchored in one of its stated complexity results.
The strongest prospective mirror is nevertheless clear, and I would lead with Theorem 4 as the mathematical anchor—not as a valid complexity anchor. Theorem 4, proved in this paper, says that with goods with copies, every MNW and complete MNW allocation is \(1/2\)-EF1WC and Pareto optimal. This is the most population-friendly part of the paper because “goods with copies” already describes repeated resources and repeated demand patterns.
Consider a large course-allocation or shift-assignment system. There is a finite set \(T\) of student or worker types. A type \(t\) is a complete additive valuation vector over course or shift categories, including eligibility information; \(\mu_t\) is the fraction of the population of that type. Each category \(g\) has a normalized supply \(q_g\), representing copies per capita. Every individual still receives an indivisible bundle \(B\) containing at most one copy from each category.
A continuous allocation is a collection
\[ x_{t,B}\ge 0, \]
where \(x_{t,B}\) is the mass of type \(t\) receiving the integral bundle \(B\). It satisfies
\[ \sum_B x_{t,B}=\mu_t \]
for each type \(t\), and
\[ \sum_{t,B:g\in B}x_{t,B}\le q_g \]
for each category \(g\). A complete version replaces the relevant inequalities by equalities. The objective is the exact high-multiplicity analogue of MNW: first maximize the mass receiving positive utility, then maximize
\[ \sum_{t,B:v_t(B)>0}x_{t,B}\log v_t(B). \]
Equivalently, after clearing rational denominators, this maximizes the product of positive utilities of a finite population of clones. The output is a mass assignment over integral bundles, not a fractional bundle for an individual.
The continuous question would be:
Given rational type masses, additive valuations, copy supplies, and optionally completeness, compute a continuum-MNW allocation \(x\). Determine whether every optimizer is support-wise \(1/2\)-EF1WC and population-Pareto-optimal, where the latter means that no feasible reassignment can weakly improve every unit of type mass and strictly improve a positive-measure set.
EF1WC should be imposed support-wise: for every used state \((t,B)\) and every used state \((u,D)\), type \(t\) must either value \(B\) at least half as much as \(D\), or be able to remove one category-copy from \(D\) that type \(t\) does not itself possess and obtain the \(1/2\)-EF1WC inequality. This avoids the atomless loophole in which exceptional individuals disappear into measure-zero sets.
The regime is plausible: thousands of students, nurses, or employees form a small number of recurring valuation and eligibility types, while course sections, shifts, or housing units are replicated. Both population and resource supply scale together. If all masses and supplies are rational, denominator clearing recovers a finite high-multiplicity instance, so this is an extension with strong clone fidelity. The authors should recognize it as their MNW/goods-with-copies problem with named individuals replaced by repeated types.
I would expect the unrestricted computational version to be Class B: hardness should transfer because the combinatorics remain in indivisible bundles, goods, and feasibility constraints, not in population multiplicity. The continuous formulation may become easier when \(|T|\) is genuinely small, but exact MNW pricing still involves choosing bundles, and the paper itself flags computation as likely difficult. Natural follow-up questions are whether fixed \(|T|\) admits a configuration-LP or approximation algorithm, and whether the \(1/2\)-EF1WC guarantee survives for every optimizer in the irrational-mass limit.
The weakest point is decisive: this is not a mirror of a computational theorem in the submitted paper. It is a faithful, author-recognizable computational extension of Theorem 4. Under the programme’s strict screening rule, that means the paper should receive “no anchored continuous mirror,” despite having a promising population-continuization direction.
The strongest negative case is source-level and decisive: this paper has no eligible computational anchor. Theorems 1–6 and Corollary 1 are fairness, efficiency, and impossibility guarantees; Theorem 7 is an imported equilibrium-existence theorem; and Theorem 8 is an existence result supported by a decomposition procedure, not a running-time or complexity theorem. Section 5 explicitly says computation is not studied and only conjectures that MNW may be hard. Under ChoCo’s named-result rule, there is therefore no computational result here to continuize.
The proposed Theorem 4 mirror is nevertheless a sensible high-multiplicity model. Repeated student or worker types, per-capita copy supplies, and mass over indivisible bundles preserve rational-clone fidelity. That is not where the objection lies.
The problem is that the proposed computational question is new. For rational masses, clearing denominators produces an ordinary finite goods-with-copies instance, and Theorem 4 already guarantees that every MNW optimizer is \(1/2\)-EF1WC and Pareto optimal. Thus checking the theorem’s guarantee contributes no new computational question. Computing the continuum-MNW optimizer is potentially interesting, but it is precisely the computation the paper does not formulate or study; it is a follow-up problem inspired by Theorem 4, not a mirror of it.
The proposed repairs—support-wise EF1WC, population Pareto optimality, irrational masses, or richer type descriptions—make the extension more defensible, but also move it farther from a direct mirror. They add new semantic and computational choices rather than recover a missing result. Replacing the indivisible bundles by fractional allocations would instead invoke the paper’s best-of-both-worlds material, but that is outcome-space continuity, outside ChoCo’s scope.
Changing the application to courses, shifts, housing, or replicated jobs cannot repair this defect: it can show that a high-multiplicity version is sensible, but cannot turn an existence theorem into a computational anchor. So the strict verdict is “no anchored continuous mirror.” The broader claim that no worthwhile research problem could ever be built from this paper is not honestly provable; the proposed MNW computation may be a worthwhile new project. It simply is not a continuous mirror of a computational result in this paper.
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.