| paper | d4-conf-aaai-2014-0452 |
| 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 compares optimal matching outcomes; it does not state an algorithm, a complexity bound, or hardness. The paper’s NP-completeness statement is unnumbered and attributed to prior work, so it cannot supply the required named computational anchor.
fails bit a — no named computational result to mirror
The proposed mirror addresses Proposition 1’s asymptotic gain; it does not cover the paper’s experiments, clearing heuristic, or runtime claims.
The strongest positive lead is Proposition 1, proved in this paper, though its proof is omitted for space. In the paper’s random compatibility model, a fraction \(\lambda\) of pairs are easy to match, with incoming-edge probability \(p_L>0\) constant; the rest are highly sensitized, with \(p_H=c/n_K\) for \(c>1\). If the kidney pool has \(n_K\) pairs and \(\beta n_K\) altruistic kidney donors, the liver pool has \(n_L=\gamma n_K\) pairs, and \(\beta,\gamma>0\), then with probability approaching \(1\), any maximum-cardinality clearing of the combined pools matches \(\Omega(n_K)\) more pairs than the aggregate of maximum-cardinality clearings of the separate pools. The clearing permits cycles up to a fixed length and chains of arbitrary length. The gain comes from altruist-started kidney chains extending through the liver pool.
There is an important qualification: Proposition 1 is a named theorem about optimal matching performance, but it does not assert a complexity classification. The paper’s statement that clearing with fixed \(L>2\) is NP-complete is unnumbered and attributed to Abraham, Blum, and Sandholm (2007). So there is no numbered complexity result here that meets the requested anchor rule. Proposition 1 can support a positive case for a continuous *matching* question, but it cannot by itself anchor a claim about whether discrete hardness dissolves.
The most credible setting is a large exchange registry like the paper’s nationwide scenario: roughly \(10{,}000\) kidney pairs and an estimated \(800\) liver pairs at steady state. The paper’s theoretical model groups pairs into a small number of sensitivity classes, with altruistic kidney donors as a separate class. That is a genuine many-agents-to-few-types regime. A real implementation would need types to preserve every trait relevant to compatibility; whether exact HLA, age, and weight profiles still give moderate support is an open practical question. The authors should recognize this mirror: it keeps their candidates, organ-specific compatibility, altruist-started chains, bounded cycles, and maximum-cardinality objective.
Call the proposed problem Continuum Multi-Organ Clearing Gain. An instance gives kidney-pair types \(T_K\), liver-pair types \(T_L\), their masses \(\mu_K,\mu_L\), a mass \(\beta\) of altruistic kidney donors, a directed compatibility law between pair types, and a fixed cycle cap \(L\). A pair type records the complete candidate–donor profile the model uses. A solution assigns nonnegative mass to feasible chains and cycles. The total mass of structures using any type may not exceed that type’s supply; altruist mass may start chains but cannot be reused. The objective is the mass of candidate–donor pairs matched. The question is whether the optimal joint clearing exceeds the sum of the separate kidney and liver optima by a positive amount. In the paper’s scaling, the continuous target is a gap of at least \(\delta>0\) after normalizing by \(n_K\), with high probability under the stated compatibility-generation law.
I would tentatively expect a Class A result for a finite-type fractional allocation formulation: its variables are masses of feasible structures, and an LP or flow formulation may compute the optimum. But that expectation depends on preserving the paper’s sparse compatibility regime. In particular, \(p_H=c/n_K\) tends to zero; replacing it with an ordinary fixed type-to-type compatibility probability can erase the sparse paths that drive Proposition 1’s gain. A useful next question is whether a sparse continuum model can retain those paths and still admit an efficient pricing or flow algorithm. One would also need to establish a discrete-to-continuous dictionary: when does the normalized optimum of the finite random exchanges converge to the mass optimum?
The weakest point is that sparse graph topology may carry information that a distribution over a few types cannot retain. The type-level LP could invent feasible chains or lose the local structure needed to find them. If preserving that structure requires one type per named pair, the high-multiplicity case has disappeared. So this is a plausible, authors-recognizable mirror of Proposition 1’s multi-organ clearing question, but the sparse-limit formulation is the key unresolved step. It covers that asymptotic matching result, not the paper’s dynamic experiments, clearing heuristic, or runtime claims.
Proposition 1 is a weak ChoCo anchor: it proves a probabilistic performance gain, not an algorithmic complexity result. Its claim is that the joint optimum exceeds the separate optima by \(\Omega(n_K)\) with high probability. The proponent’s normalized target, a positive gap \(\delta\), is essentially that same asymptotic claim after division by \(n_K\). Asking how to compute a continuous optimum would be a new problem, but Proposition 1 gives no computational result to carry over. The paper’s NP completeness statement concerns earlier work.
The proposed finite type LP also loses the feature that drives Proposition 1. For highly sensitized pairs, an edge has probability \(c/n_K\): the edge density tends to zero even as each pair has expected degree \(c\). The gain comes from long paths in that sparse realized graph. A limiting compatibility matrix between types records zero for those edges and loses the paths; a matrix of edge probabilities records averages, not which disjoint chains actually exist. To preserve the realized chains, a type must encode its compatibility neighborhood. That produces typically \(\Theta(n_K)\) distinct realized types, so the support no longer offers high multiplicity.
A more exact continuum model could use a sparse random network limit or a graphing. That retains the topology, but changes the input from a distribution over a small set of types to a network carrying individual adjacency structure. The question then becomes how to optimize chain and cycle packings on that network. Proposition 1 already establishes its stated asymptotic gain; the new computational question is not a high multiplicity algorithm for the paper’s exchange instance.
There is a credible nearby possibility: group pairs by complete clinical profiles and make compatibility deterministic by type. A mass version of clearing on that finite type graph could be well posed and might admit an LP. But it removes the paper’s sparse random compatibility mechanism, so it does not rescue the proposed mirror of Proposition 1’s result. It would be a mirror of the broader clearing problem under a different regime. That is where the negative case is weak: I can reject this anchor as a useful continuation of the paper’s main theorem, but I cannot honestly rule out a worthwhile continuous model of organ exchange more generally.
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.