| paper | d4-conf-aaai-2023-0087 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | yes |
The paper gives a coherent integer programming method and reports experiments, but no numbered theorem, lemma, corollary, or proposition stating a computational result. Theorem 1 is a counterexample to guaranteed existence, while the runtime results are empirical and the cited hardness result is prior work. The proposed mass-allocation problem is a plausible follow-on, but it has no qualifying result in this paper to anchor a continuous mirror.
fails bit a — no named computational result to mirror
No result qualifies as an anchored computational mirror. The IP formulation and experiments compute stable outcomes and maximize matched children on five datasets; Theorem 1 shows stable outcomes need not exist. The proposed mass problem covers the paper’s stability and matching objective, but remains an unanchored follow-on.
On the required anchor rule, this paper has no qualifying named computational result. Its only numbered result is Theorem 1, proved here: a stable outcome may fail to exist even with four children and two daycares. That is an existence counterexample, not a complexity result. The paper’s IP model and its experiments—stable outcomes found for all five datasets, in at most two seconds—are relevant evidence, but they are not a numbered theorem about an algorithm’s complexity. The NP-hardness result for couples that the paper cites is prior work, not a result proved in this paper.
The strongest positive case I can make is therefore a proposed follow-on question, not a mirror anchored in a named computational result. Call it Family-Stable Daycare Allocation\(_\infty\). An instance has a finite set of family types \(T\), rational masses \(\mu_t\) summing to \(1\), daycares \(D\) with rational capacities per unit population, and a dummy daycare \(d_0\). Each type specifies the complete family profile the problem uses: its children’s grades and individual preferences, initial assignments, the family’s ranking over daycare tuples, and its daycare priority categories. Let \(k_t\) be the number of children in type \(t\), and let \(\omega_t\) be its initial tuple.
The decision variables are \(x_{t,a}\ge 0\), the mass of type-\(t\) families assigned tuple \(a\in D^{k_t}\). Require \(\sum_a x_{t,a}=\mu_t\), permit only family-rational tuples \(a\succeq_t\omega_t\), and enforce each daycare’s capacity. The objective is to maximize matched children:
\[
\sum_{t\in T}\sum_a x_{t,a}\bigl(k_t-\#\{i:a_i=d_0\}\bigr).
\]
For stability, forbid any positive mass \(\varepsilon\) of families of type \(t\), currently assigned \(a\), from moving together to a preferred tuple \(b\succ_t a\) when every daycare in \(b\) can accommodate those children alongside the incumbent children with higher priority than at least one incoming child. The solution is either “no stable mass assignment exists” or a stable assignment \(x\) attaining the maximum.
The regime I have in mind is a large municipality assigning many families across repeated annual cohorts, with capacities growing with the cohort and a relatively small catalogue of complete family profiles. The paper’s datasets—550 to 1,589 children across 33 to 72 daycares—make the application recognizable, but do not establish that such a small catalogue exists. The type must include preferences, initial enrollment, and priorities; counting only common family circumstances would overstate how many agents are truly interchangeable.
This formulation keeps the paper’s central problem: families choose bundles for siblings, daycares have capacity and priorities, and the objective rewards matched children subject to family stability. It asks about the underlying allocation problem, not whether its IP encoding is easy. My tentative expectation is hardness transfers: the combinatorics in unrestricted sibling bundles and priorities may survive aggregation. That is conjecture, not a result of this paper; fractional assignment of a type could also change existence and complexity. The cited hardness result for couples makes the guess plausible, but does not establish it here.
The weak point is the type regime. In the paper, family rankings can be long and individualized, and daycare priorities are strict down to tie-breaking details. Treat those details as part of type and the support may grow almost as fast as the number of families. Keeping support small likely requires priority categories with exchangeable tie-breaking, a co-modification the authors might accept as a continuum analogue but that the paper does not study. So the allocation question is concrete and plausibly theirs; the paper supplies no named computational anchor for it, and its data do not yet show that the high-multiplicity regime is a good fit.
The proponent has no qualifying anchor. The paper’s only numbered theorem, Theorem 1, shows that a stable outcome need not exist; it says nothing about computational complexity. The cited NP-hardness for couples is prior work, and the paper’s own IP results are empirical: they report runtimes on five datasets, not a worst-case bound or a complexity classification. By the programme’s stated test, there is no named computational result here to mirror.
Family-Stable Daycare Allocation\(_\infty\) is a coherent follow-on question, but it is a new complexity problem, not a continuous version of a computational result in this paper. Its high-multiplicity case is also unsubstantiated. A type must include the family’s full tuple ranking, initial assignment, and the children’s priorities at each daycare. The paper uses strict daycare rankings, including tie-breaking; preserving exact priority positions makes otherwise similar children distinct types. Replacing those rankings with shared priority classes could create repeated types, but it changes the stability model, and the paper gives no evidence that the municipalities would regard that change as faithful. Its datasets have hundreds to over a thousand children, but do not show that the number of complete profiles is small; some family preference lists run to hundreds or over a thousand tuples.
There is a plausible route to a continuum model: repeated family profiles, priority tiers, and capacities that scale with the cohort. But the paper neither establishes that regime nor identifies a computational bottleneck it would address. Its IP found stable outcomes for all five datasets in at most two seconds. That is not a worst-case guarantee, but it leaves the proposed mirror without an evident practical motivation as well as without a paper result to anchor it.
I cannot honestly conclude that no scenario could make the continuum allocation question worthwhile. A municipality with many families sharing complete profiles could provide one. The negative case is narrower: this paper does not establish that scenario, and its proposed continuous allocation problem should not be presented as a mirror of one of the paper’s named computational results.
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.