| paper | d4-journals-mss-2021-0016 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | no |
Theorem 1 and Theorem 2 characterize representation and explicit description; neither states an algorithmic result, a complexity bound, or a hardness claim. The proposed finite-support mean comparison is statable, but it computes a different aggregate and mirrors neither theorem, so no continuous computational question remains anchored in this paper.
fails bit a — no named computational result to mirror
Theorem 1's representation characterization and Theorem 2's explicit-description characterization are left uncovered; the proposed mean comparison reproduces neither theorem's finite-versus-infinite boundary.
The strongest positive angle I can defend is to treat generations as the population. In a long-horizon setting, many generations could share the same one-period utility level. A continuous version would represent them by a distribution \(\mu\) over utility types \(y\in Y\), where \(\mu_y\) is the fraction of generations of type \(y\). For bounded utilities, one concrete problem is Continuum Generation Welfare Comparison: given two finite-support rational distributions \(\mu,\nu\), decide whether
\[
\sum_{y\in Y}\mu_y y \;\geq\; \sum_{y\in Y}\nu_y y.
\]
The solution is the comparison of their aggregate utilities. This has a connection to the paper’s construction: when a stream’s empirical distribution converges, its liminf average utility is the average under that distribution.
But this is only a weak modeling analogy, not a valid computational mirror of a named result. Theorem 1 says that a real-valued social welfare function satisfying anonymity and asymptotic density-one Pareto exists on \(Y^{\mathbb N}\) iff \(Y\) is finite; its finite-domain existence result is cited to Petri (2019), and the paper establishes the infinite-domain impossibility. Theorem 2 says that a constructive social welfare order with those axioms exists iff \(Y\) is finite, using the paper’s result that such an order on the infinite domain implies a non-Ramsey collection. Both are representation or constructiveness theorems, not complexity results about an encoded computational problem. In particular, “constructive” here does not mean an efficient algorithm.
The continuous comparison problem is easy to state and compute, but it discards features central to those theorems. A distribution forgets the ordering of generations and cannot represent every stream; the original anonymity axiom only identifies streams under finite coordinate swaps, while a continuum distribution identifies much more. And for bounded \(Y\), the integral gives a representation even when \(Y\) is infinite, so it does not preserve the paper’s finite-versus-infinite boundary. I would expect this comparison problem to be tractable, but I would not claim the authors or a referee should accept it as the continuous form of their problem.
So the honest verdict is that the paper has no named computational anchor for a ChoCo-style mirror. Theorem 1 is the closest result to anchor on, but under the requested standard it is not eligible. The continuum-generation model is the best positive intuition I can offer; its weakest point is also decisive: it turns an axiomatic question about ordered infinite streams into average-utility computation over distributions.
The strongest case against a ChoCo mirror is that neither Theorem 1 nor Theorem 2 states a computational result. Theorem 1 characterizes when a real-valued welfare representation exists on the full space \(Y^{\mathbb N}\). Theorem 2 characterizes when an order has an explicit description; its obstruction is the existence of a non-Ramsey collection. Here “constructive” concerns describing a pairwise ranking, not an algorithm’s running time. Neither theorem supplies encoded instances, an optimization or decision problem, or a complexity measure. Restricting streams to a chosen input format and asking for an algorithm would add the computational problem the paper itself never formulates.
The proposed distribution comparison does not repair that gap. The paper’s objects are utility streams indexed by generations. Anonymity permits finite swaps, while asymptotic density-one Pareto asks where along the ordered sequence improvements occur. A distribution over utility levels records frequencies, not that arrangement; many streams have no limiting empirical distribution at all. If types retain enough information to recover each generation’s position, the proposed mass aggregation no longer compresses the temporal structure that drives the axioms.
The best repair would replace natural density over generations with full measure in a continuum of generation-times, and replace anonymity with invariance under measure-preserving rearrangements. That is a coherent welfare-theoretic question, but it changes the paper’s temporal model and remains a question about axioms and representation. Comparing integrals for rational simple functions is computationally easy; it does not mirror either theorem’s finite-versus-infinite domain boundary or non-Ramsey construction.
The suggested average-utility comparison has a more direct mismatch. On \(Y=[0,1]\), let \(y_n=0\) and \(x_n=1/n\). Every generation is strictly better off in \(x\), so the paper’s axiom requires \(x\succ y\). Yet their limiting average utilities are both zero. Restricting inputs to finite-support distributions avoids this example by narrowing the stream class, but then the comparison answers a new, restricted aggregation question rather than the paper’s existence result over all streams.
So the negative case is strong on ChoCo’s terms: there is no named computational anchor to continuize, and the most natural population reformulation discards or replaces the index structure central to both results. That does not rule out worthwhile computational work inspired by the paper. It does make such work a new problem, rather than a continuous mirror of its theorems.
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.