Overview
Eval runs
Section titled “Eval runs”Eval-run data plane (start, list, get, cancel, events SSE). Dispatches an eval run against a registered suite. Dispatch is per-EvalKind; kinds without a registered dispatcher return 422 dispatcher-not-registered. result is kind-specific opaque JSON on the wire — for accuracy it is { passCount, totalCount, meanScore, perCase[] }. SSE events mirror run streaming: per-case progress followed by a terminal frame carrying the aggregate. Caller-plugged via EvalRunBinding.