Jev Fundamentals
A free course on Jev (TypeSafe AI): the model that gives up text generation and returns typed, calibrated decisions your code can use directly.
Free to read · unofficial educational resource; Jev and TypeSafe are trademarks of TypeSafe AI · vocabulary in the GlossaryS01E01: Jev - The Model That Answers Instead of Writes
Read a Jev request and answer as fluently as a function signature: the three question primitives (Noul, Choice, Score), request and answer anatomy, fan-out economics, and a worked rubric-to-questions example.
S01E02: Jev State and Typed Questions
Author the two things you write by hand in every Jev request — the state and the questions. The three question types up close, criteria that define your answer space, the vendor's wording failure modes with the repeat-run procedure, independence by design, and a complete worked request.
S01E03: Confidence, Calibration & Composite Scoring
What Jev's confidence numbers promise, threshold routing, uncertainty gates for torn answers (margin, low confidence, torn distributions), composite scoring with weights in code, measuring calibration instead of trusting it, and calibration's payoff: the sweep and the band.
S01E04: Pins and Receipts (Keeping Your Setup Stable)
Why the model name you send is not a promise: aliases resolve at request time and move, hosted limits shift, aggregators fall back, a platform router changes who answers. Pin the versioned build and log a receipt from every call, across every venue, on one wire contract.
S01E05: The Local Benchmark (Tying It All Together)
A pre-registered benchmark of five local decision models against the hosted reference: five task chapters walk one real item and one response schema at a time, then the whole-run scoreboard with the tie rule, five context walls of five kinds, the recorded-skip discipline, and the labeler problem, closing with the developer's decision branches and two sample routing policies. 15-question quiz.