Jev Gives You Probabilities, Not Decisions (Olivier Miossec)
External DEV write-up by Olivier Miossec (posted 2026-10-09, edited 2026-10-25): the mental model behind hosted TypeSafe Jev—state vs questions, Score / Noul / Choice, confidence thresholds, and writing atomic questions—with PowerShell examples that apply in any language.
What this guide points to
Olivier Miossec published Jev Gives You Probabilities, Not Decisions on DEV (posted 2026-10-09, edited 2026-10-25). It follows his cloud-ops PowerShell article and explains the concepts: TypeSafe’s System One inspiration, why Jev returns typed probabilities instead of prose, how state (data only) differs from questions, and how Score, Noul, and Choice answers should drive code-owned thresholds.
Worked ideas in the post:
- State hygiene — prefer JSON objects; strip IDs and empty fields; keep the 32k context budget in mind; non-English text may be less accurate.
- Score / Noul / Choice — ordered levels, yes-probability, and closed options (add an
unclearescape for Choice). - Confidence vs automation — concentrated probability → automate; mid band → confirm; low → investigate. TypeSafe’s 0.9 / 0.5 sketch is a starting point; measure your own thresholds on labeled states.
- Atomic questions — one concern per question; never ask Jev what a script can decide; batch independent questions in one call.
Examples use Doug Finke’s dfinke/Jev PowerShell module against hosted TypeSafe Jev; the author notes the ideas apply to any language. This page is an unofficial handbook pointer — verify against the article and TypeSafe docs.
Takeaways
- Jev supplies probabilities; your code turns them into decisions.
- Put data in state and judgment criteria in questions; keep them separate.
- Calibrate automation thresholds on your own labeled set, especially for non-English or niche states.