| paper | d4-conf-aldt-2021-0026 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | no |
The paper reports experimental MARL outcomes on finite, individually indexed markets, but contains no numbered theorem, lemma, corollary, or proposition asserting a computational result. Its episode counts and measured stability and fairness are empirical settings and findings, so there is no eligible result to anchor a population-continuum mirror.
fails bit a — no named computational result to mirror
No paper result is covered; the experiments on stability, instability, fairness, and learned policies remain finite-market empirical findings.
I can’t make a compliant anchored positive case from this paper. It contains no numbered Theorem, Lemma, Corollary, or Proposition asserting a computational complexity result. Its contribution is an experimental study of MARL on small markets of individually indexed agents, reporting stability and fairness measurements. Those findings are not eligible anchors under your criteria, so there is no per-anchor continuous problem or complexity prediction to hand to a prover.
A plausible setting for a *different, unanchored* mirror would be a large hiring market where workers and employers fall into a small number \(\tau\) of complete preference and utility types. The masses would be type proportions, and the decisions would describe how type populations meet and match over locations. But the paper’s experiments do not establish a named computational result about that population model.
The strongest objection is that this paper offers no named computational result to mirror. Its claims are experimental: SARSA often produces stable or nearly stable, fairly rated outcomes on small markets. It gives no complexity bound, convergence guarantee, or general algorithmic stability result. The reported \(100\text{k}\)–\(400\text{k}\) training episodes are an experimental setting, not a sample-complexity result. The other analyst also concedes the gap: the hiring-market proposal is “different” and “unanchored.” It cannot supply the missing paper result.
The best version of that proposal is still a sensible new model: worker and employer types could include location, acceptability, preferences over counterpart types, utilities, and noise, with matching flows between types. One could study mean-field MARL or compute stable allocations in that setting. But preserving the paper’s indexed opponents and per-agent policies makes the type space grow with the roster; compressing opponents into categories and sharing policies creates a different environment and algorithm. Either way, the paper’s finite-market experimental findings establish no corresponding result about the continuum model. That work might be worthwhile, but it would be a new research project, not a continuous mirror of a computational result in this paper.
So the case against a ChoCo mirror is strong under the requested anchor test: no anchor was offered, and none of the paper’s experimental claims supplies the programme’s kind of computational target. The broader claim that *no scenario* could yield worthwhile continuous work is weak. A typed hiring-market model is plausible; I cannot honestly rule out its value. The defensible negative conclusion is narrower: this paper gives no anchored result for that mirror to continue.
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.