Multi-agent systems and production
Observability and cost
Production asks three questions of a pile of trace records. Thread a run id through everything, count failures in buckets that have different fixes, and report cost per completed task rather than per call.
Not yet
This one opens once you have finished Self-correction.
Progress is kept against your account, so finishing a module needs one. Everything up to here is open without.