What problem does it solve? Integrating an LLM into deterministic software workflows often produces unreliable free-text output that is hard to validate, route, and audit. This Skill guides the design, implementation, and review of TypeSafe AI Jev/System One integrations, which return structured typed judgments (Choice, Score, Noul) instead of prose, so code stays in control of calculations, validation, and execution. ## Core Features & Use Cases - Decision Design: Decide whether Jev fits a workflow, declare typed Choice/Score/Noul questions, shape state, and batch independent questions over one state. - Integration Guidance: Implement calls through the TypeScript SDK, Python SDK, HTTP API, OpenRouter, or Vercel AI Gateway with strict response validation and error classification. - Calibration & Operations: Calibrate thresholds on labeled data, run self-consistency studies, handle retries and failures, and audit Jev use in Tyche rule pipelines. - Use Case: You are building a support-ticket router. Use this Skill to declare a Choice question for routing, a Noul for urgency, and a Score for impact, then apply a versioned, evaluated policy that routes uncertain answers to human review. ## Quick Start Use the jev-system-one skill to design and validate a Jev integration that classifies incoming support tickets by route, urgency, and impact with calibrated review thresholds.
