System Health
Activity events / minute (last 2h)
Latency distribution bm25-memory hook
Utility rate share of injected items Claude actually referenced ⓘ how
Quality notices rate-based, informational
Other signals
Retrieval distribution node type · query type · retrieval method — schema v1.6
Knowledge graph decisions · docs · prompts — coral = selected prompt, edges = retrieval + topic + temporal
Function proof — real samples recent prompts → actual hook outputs (eye-check relevance)
Per-prompt utility pill:
≥50%
≥25%
<25%
—
= how much of what CTX surfaced was actually used in Claude's response.
Blocks below show what CTX surfaced into context. A filled block with 0% utility = CTX injected items
but Claude didn't reference them — honest signal, not a bug; usually means the prompt was about a different topic than what BM25 matched.
— = the reply isn't persisted yet (lands on Stop hook ~30s after the turn); score fills in on next refresh.
CM entries are project-scoped history, not this prompt's reply.