The basics
What is RLCD?
RLCD stands for Reinforcement Learning for Calibrated Decisions. It is TypeSafe's way of training AI to make specific choices and give useful odds for each answer. Instead of writing freely, Jev gets some context and a question with known possible answers. The aim is for its probabilities to match how often it is right over many similar questions.
What does calibrated mean?
A model's probabilities should line up with reality over many tests. If it says 70% a hundred times, about 70 of those answers should be right. That says nothing certain about any one answer.
What is Jev?
Jev is TypeSafe's decision model. Give it some context and a question with limited possible answers. It returns a value and probabilities, not a paragraph.
What kinds of questions can it answer?
It can answer yes or no (Noul), choose from a list (Choice), or give a rating (Score). A support queue might ask whether a ticket is urgent, which team owns it, and how risky it is.
Why chain several decisions?
Instead of asking one model to handle a whole workflow, you ask a few narrow questions. Your code combines the answers and decides when a person should step in.
Can Jev be wrong?
Yes. A tidy output and a confident number do not make an answer true. Check it on your own examples and keep a human review step where mistakes are costly.
Go deeper: TypeSafe introduction · Decision types · Launch explanation