Glossary
Short definitions for the terms that show up in Jev requests and answers.
Glossary: System One
System One is TypeSafe's class of models for fast structured decisions. Jev is the first public instance.
Glossary: jev-latest
jev-latest is the SDK-default alias for the current official Jev release. It can move when TypeSafe ships.
Glossary: RLCD
RLCD is TypeSafe's name for Reinforcement Learning for Calibrated Decisions — named, not fully specified, in public docs.
Glossary: Choice
Choice is Jev's unordered classification primitive: up to 255 options, a winner, a full distribution, and confidence.
Glossary: Score
Score rates state on 2–10 ordered levels and returns a possibly fractional expected position plus confidence.
Glossary: Noul
Noul is Jev's yes/no primitive: P(yes) in [0, 1] with no separate confidence or probabilities map.
Glossary: Confidence
Confidence is a 0–1 summary of how peaked a Choice or Score distribution is. Noul answers have none.
Glossary: State
State is the text or JSON payload every question in a Jev request evaluates. Images and audio are not accepted.
Glossary: Intent
Intent is a closed label you ask Jev to assign so code can route to a handler — not a generated plan.
Glossary: Signal
A signal is a typed Jev answer your code can consume — noul, choice+probabilities, or score — not a generated insight sentence.
Glossary: Action space
Action space is the closed set of acts your code may take after Jev answers — including abstain. Not a generated plan.
Glossary: Verdict
A verdict is the action your code commits after Jev answers — allow, hold, block — not a generated essay or official API field.
Glossary: Decision contract
A decision contract is the closed questions, criteria, and code thresholds you ship — not a TypeSafe legal term.
Glossary: Schema validation
Schema validation is TypeSafe's 422 on a bad body, plus your closed criteria — not proof the answer is factually correct.
Glossary: Threshold
A threshold is a numeric cut your code applies to confidence, noul, or score — TypeSafe examples are illustrations, not defaults.
Glossary: Confidence threshold
A confidence threshold is your act/review/abstain floor on Choice or Score confidence — docs illustrations are not defaults.
Glossary: Calibration
Calibration means higher confidence should mean higher accuracy. TypeSafe claims RLCD trains for this; you still measure it.
Glossary: Abstention
Abstention is a first-class outcome: low confidence, other, or a Noul near 0.5 — your code refuses to auto-act.
Glossary: Fallback
A fallback is the safe path your code takes when Jev is unsure, down, or out of policy — not a second invented label from the model.
Glossary: Guardrail
A guardrail here is a typed Jev question plus your threshold — not a certified security product or TypeSafe firewall.
Jev glossary: Policy check | TypeSafe Jev System One
Define policy check in the Jev decision model — Jev policy check definition. Independent unofficial notes; confirm against docs.typesafe.ai.
Glossary: Ranked route
A ranked route is an ordered list of handlers from Jev probabilities — code still applies floors before it follows rank 1.
Jev glossary: Human handoff | TypeSafe Jev System One
Define human handoff in the Jev decision model — Jev human handoff definition. Independent unofficial notes; confirm against docs.typesafe.ai.
Glossary: Decision trace
A decision trace is the logged state, questions, answers, model id, and policy branch — not a generated chain of thought.
Glossary: Audit trail
An audit trail is the durable record of state, answers, model pin, and verdict — Jev does not write a reasoning essay to store.
Jev glossary: Evidence bundle | TypeSafe Jev System One
Define evidence bundle in the Jev decision model — Jev evidence bundle definition. Independent unofficial notes; confirm against docs.typesafe.ai.
Jev glossary: Retrieval grounding | TypeSafe Jev System One
Define retrieval grounding in the Jev decision model — Jev retrieval grounding definition. Independent unofficial notes; confirm against docs.typesafe.ai.
Jev glossary: Risk band | TypeSafe Jev System One
Define risk band in the Jev decision model — Jev risk band definition. Independent unofficial notes; confirm against docs.typesafe.ai.