◆ LLM-written

KB retrieval observability

how effectively the knowledge index serves agents · built 2026-08-27 08:00 MST · ARCH-121 · hover any metric label for what it measures and what events feed it

Retrieval quality golden set · 44 cases · k=10

36%
recall@1
△ below target
50%
recall@5
△ below target
55%
recall@10
△ below target
0.43
MRR
△ below target

by query style

stylen r@1r@5 recall@10MRR
keyword3244%53%56%0.483
nl1217%42%50%0.29

by doc kind

kindn recall@10MRR
agents40%0.0
architecture250%0.056
decisions2479%0.659
reference20%0.0
runbooks1136%0.273
skills10%0.0

No misses — every golden query lands in top-k.

eval generated 2026-08-27 07:48 MST

Live queries

No query traffic logged yet. kb-queries.jsonl fills as co sessions cycle onto the instrumented server.

Materialization coverage

No materializations logged yet. kb-materializations.jsonl fills as knowledge-ctx use runs on session start.