Summary
Retrieval routing should not treat a single analyst/BI-classified query-history observation as enough to make query history the primary evidence route. This matters for intermediate, low-volume, and infrequently used tables where query history may exist but is too thin to be reliable.
Stack
- Branch: stack/query-history-route-threshold
- Base: stack/documentation-route-strength
- Parent PR: fanduel/fd-knowledge-graph#8
Expected behavior
- Query history routes as primary only when analyst/BI usage exists and analyst_usage_evidence_count is at least 2.
- Thin analyst/BI query-history evidence remains visible in retrieval context.
- Thin query-history-only tables stay review-first with a routing-threshold warning.
Safety
No raw SQL or user identifiers are introduced. The change uses graph-safe counts already present in retrieval signals.
Verification
- ruff check .
- pytest -q
- git diff --check
Summary
Retrieval routing should not treat a single analyst/BI-classified query-history observation as enough to make query history the primary evidence route. This matters for intermediate, low-volume, and infrequently used tables where query history may exist but is too thin to be reliable.
Stack
Expected behavior
Safety
No raw SQL or user identifiers are introduced. The change uses graph-safe counts already present in retrieval signals.
Verification