Why it is a question
THEORY-324 now states Ashby’s necessity for a system that adapts by trial and error. Under the trial-and-error rule, the parameters must stay put when the essential variables are within limits and change when they are not, so the parameters depend on those variables, and a channel follows from the rule itself. The former promote_when asked more: that any system which comes to keep its essential variables within limits, against an environment whose coupling to them it does not know, carries such a channel.
In that wide form it fails trivially for a regulator whose single fixed setting is stable in every environment of the class: nothing needs to change, so nothing needs a channel. The question excludes that case. What remains open is whether a system can do without the channel when it does need different settings in different environments, but does not follow the trial-and-error rule. One example is a regulator that learns the environment’s coupling from the environment’s own variables. Another is one whose settings are switched by a cue that happens to predict which setting is needed.
What would answer it
A proof that, whenever different environments in the class need different settings, the parameters’ transitions must depend on the essential variables (directly or through intermediates), would answer it yes. A system meeting Ashby’s assumptions that finds the needed setting with no such dependence would answer it no, as would one that relies only on a cue correlated with the environment and carrying no information from the essential variables. More demonstrations that ultrastable machines adapt do not settle it: they show sufficiency only.