Guides
Recipes
Reusable pipeline patterns for combining Jev with generative LLMs.
Recipe: Cascade Routing — Jev Coarse, Jev Fine, LLM Last
The front-filter pattern scaled: Jev coarse-screens everything, a finer Jev pass refines survivors, and an expensive LLM sees only what deserves generation. The cost math that makes it obvious.
Recipe: Confidence Gating — Automate the Sure, Escalate the Rest
Use Jev probability and confidence thresholds to auto-execute high-certainty verdicts and fall back to humans or LLMs on low-certainty ones. Three-zone design with working code.
Recipe: The Jev → LLM Front-Filter Pipeline
The foundational Jev recipe: put a cheap judgment model in front of an expensive generative one. Pattern, judgment design, routing rules, and cost mechanics.
Recipe: Speculative Fan-Out — Many Questions, One Call
Pack multiple small Jev questions into a single request so one state transmission returns N typed verdicts in parallel. The cheapest latency and cost optimization in the Jev toolkit.