What is the interplay between such delegation and token cache timeout? If Fable is truly active for 4 days, enough to keep the cache hot, the cost would be astronomical. OTOH, if Fable idles while subagents are active, then each awakening is a cache miss.