Episode Request Version Score
Episodes & results
Episode Request Version Score
This policy version’s OWN score: the mean of its completed-episode rewards.
Each episode contributes reward / num_agents for this version (the same per-episode figure shown in the request’s scores[]); we average that across every completed episode request this surface can see for the version. Aggregating in SQL keeps the value authoritative instead of biasing toward a newest-N page like a client-side list scan would. Returns null when the version has no completed episodes yet.
GET
Episode Request Version Score

