- Issue 1: Normalize domain lookup to lowercase on both sides (flow category names and session problem_domain) to fix case-sensitive mismatch in coverage endpoint
- Issue 2: Count distinct ai_session_id (not PsaPostLog.id) in doc_pushed funnel step to avoid inflating counts on retried sessions
- Issue 3: Clamp recency_score to [0.0, 1.0] with min/max to handle negative days_since from future timestamps (clock skew/test data)
- Issue 4: Wrap func.sum(case(...)) results with int() in dashboard endpoint to handle Decimal returns from asyncpg
- Issue 5: Remove dead domain_flow_counts_result query (result fetched but never consumed) to eliminate unnecessary DB round-trip
- Issue 6: Replace select(Tree) with select(Tree.id, Tree.name, Tree.tree_type) in flow-quality endpoint to avoid loading large tree_structure JSONB column
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
chihlasm
94fbb38f84