Curious about graceful degradation β if one of the underlying models
Fugu is coordinating hits a rate limit or goes down mid-task, does the
whole request fail, or does it reroute to an alternative model
transparently? That failure-mode behavior seems just as important as
the happy-path orchestration story for anything running in production.