Mission: TypeSafe / Jev

Ship AI-powered software where code owns the workflow and Jev supplies narrow, typed judgments (routing, gates, scores) for agent loops, CI, and product paths — without treating a chat LLM as the control plane.

Success looks like

• Correctly choose Choice vs Noul vs Score for a real judgment, and say why
• Design a request: structured state + atomic questions + confidence/probability gates in code
• Wire one end-to-end decision (e.g. PR triage or agent-output gate) and validate thresholds on sample data
• Know when not to use Jev (generation, open-ended reasoning, agent planning)

Constraints

• English lessons; fatherly grammar corrections in chat
• Prefer live docs.typesafe.ai over model memory
• API key already on this machine; keep credentials out of chat and lessons
• Parallel track: loop engineering — keep this workspace focused on System One

Out of scope

• Training RLCD / model internals beyond what the primer needs for intuition
• Replacing coding agents with Jev
• Full product redesign of Wellhub systems (use them only as optional practice targets)