| paper | d4-journals-jet-2014-0032 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | no |
None of the named results asserts that an object can be computed or that computing it is hard. The proposed equilibrium-computation project would need a specified computational target and input representation that the paper does not provide, so it cannot supply the required computational anchor.
fails bit a — no named computational result to mirror
No named result supports a computational mirror; Propositions 1, 9, and 10 are existence and asymptotic learning results, and the paper’s other results do not supply a computational anchor either.
My strongest honest case is only a weak setting-level fit, not an eligible computational mirror. A large adoption cohort choosing between technologies could plausibly be high-multiplicity: agents share payoffs and signal and sampling rules, while their private signals and information about position distinguish their types. Mass would mean the fraction of agents of each such type, and the outcome of interest could be the fraction choosing the superior technology. The paper itself studies what happens as the number of agents grows.
But the paper has no named computational result to anchor a continuous computational question. Its closest named results are analytical: Proposition 1, proved here, establishes equilibrium existence for each finite \(T\); Proposition 9 establishes social learning in every equilibrium sequence with unbounded signal strength and stationary sampling; Proposition 10 establishes constrained-efficient learning under bounded signal strength and stationary sampling. These concern existence and asymptotic learning, not computational complexity or algorithms.
One could imagine asking for an equilibrium action rule in a continuum adoption model, with mass over position and private-information types and a sampling rule over earlier mass. Yet that would add a computational question the paper does not formulate, while its central results concern the ordered sequence and its large-\(T\) limit. Under the requested anchor rule, I therefore have no valid per-anchor continuous problem to hand to a prover. The population-scale setting is plausible; the case for a continuous computational mirror of this paper’s named results is not.
The strongest case against is that the three anchors are equilibrium and learning theorems, not computational social choice problems. Proposition 1 proves equilibrium existence; Propositions 9 and 10 characterize asymptotic learning. None gives an algorithmic problem with an encoded input and output—such as finding a manipulation, optimizing a campaign, or deciding winner robustness. Making the population continuous does not supply that missing computational task.
For Proposition 1, the best proposed mirror is to compute an equilibrium of a continuum adoption game. But the paper allows arbitrary signal distributions, sampling rules, initial histories, and position information. It does not specify finite encodings for those objects or a representation of the equilibrium strategy maps. A computational version would need to impose those choices—perhaps finite signal alphabets, rational probabilities, and a finite strategy table. That could make a useful equilibrium-computation project, but it would be a newly specified computational game, not a continuous version of the existence result. In the literal continuum model, strategies become functions over position and information, and the proposition remains an analytic existence claim.
Proposition 9 has the same problem. Its result is that average utility approaches its maximum as \(T\) grows. A continuous time or rank model could express that learning limit, but the proposition supplies no computational output to calculate. Asking for an equilibrium schedule or a finite-horizon convergence bound would add a new task. Proposition 10 replaces the limit with a lower bound, \(u_{\mathrm{cl}}\); that is still a statement about equilibrium outcomes, not an optimization problem over the population. Computing worst-case equilibria could be an interesting extension, but it is not an algorithmic question posed by the paper.
There is also a tension between preserving the model and getting genuine multiplicity. Agents occupy distinct positions, and a position determines which past actions may enter their sample and how their action relates to others’ positions. If type records the full role relevant to play, the \(T\) positions generally produce \(T\) types. Replacing rank by a continuum of normalized positions gives a measurable schedule of agents, not a population with moderate support over repeated types. Grouping people into large adoption cohorts is more plausible, but it changes the one-agent-per-position sampling structure: people in a cohort must share a role, and within-cohort ordering and observation must be removed or redesigned. That is a reasonable new model, but it does not preserve the named results’ setting.
This is not a proof that no worthwhile project can be built from the paper. Large adoption cohorts could motivate a high-multiplicity equilibrium-computation problem in an adjacent area. That possibility is the weak point in the negative case. But it would require choosing a new computational object and a cohort model; Propositions 1, 9, and 10 do not themselves anchor a continuous computational mirror.
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.