Same coverage_query tool, two implementation tasks:
A asks about Graphonomous MCP deployment (coverage 0.532, 4 knowledge gaps)
and gets a learn recommendation; B asks about WebHost.Systems
Supabase auth (coverage 0.558) and also needs more knowledge.
The act threshold requires coverage ≥0.720 and uncertainty ≤0.350 —
both tasks fall short, but with 74 nodes in the knowledge graph (up from 29),
coverage scores are notably higher. A hypothetical review_goal shows what
happens when the gate opens.