# Expected Diagnosis: premium-keyspace-storage-query-semantics

This ad-hoc premium case evaluates whether the agent can reason about live
metric semantics instead of blindly replaying one PromQL pattern across every
premium cluster.

The expected response should:

- inspect the single selected TiKV instance reported by the runtime window
- inspect `tikv_store_size_bytes{type="used"}` directly
- distinguish the query scope of the direct metric from the fallback metric
- avoid assuming every premium cluster exposes the same keyspace labels
- recognize whether the active premium cluster exposes keyspace-level
  dimensions for `tikv_store_size_bytes`
- recommend `tikv_engine_size_bytes{db="kv"}` on the fallback path where direct
  keyspace semantics are missing
- emit exactly one JSON code block with the required fields:
  `target_metric`, `evidence_sufficient`, `clusters`

The judge intentionally checks the final semantic contract, not the exact query
prose. The key failure mode is treating premium metric routing and keyspace
label structure as universal across clusters.
