The short version
TypeSafe exposes three AI primitives. Choice picks an option from a list. Score rates the state against a rubric. Noul asks whether a statement is true, on a 0 to 1 scale.
All three can be mixed in one request. Every question is evaluated against the same state, in parallel and in isolation.
Why this matters
Most AI failures in production are interface failures, not intelligence failures. A model asked an ambiguous question in prose gives a plausible answer in prose, and the code downstream has to guess.
Primitives force the question to be specific. If you cannot express what you want as a choice, a score or a truth estimate, the problem is the question — not the model.
Choice: bounded categories
Use Choice when the set of valid answers is known and finite: which channel, which team, which content pillar, which tone. You pass a criteria map where each key is an option and the value describes it.
You get back the chosen key, a probability for every option, and a confidence value. The probabilities are often more useful than the pick — a 0.55 / 0.44 split is a signal to escalate, not a decision to ship.
Score: rubrics instead of vibes
Use Score when the answer is a position on a scale and you can describe what the scale points mean. You supply criteria as an ordered list and get a score plus a legend linking numbers back to your descriptions.
This is how you replace “rate this post from 1-10” with something reproducible. The legend is the part that makes it auditable, because it fixes what a 2 means.
Noul: probability of a claim
Noul returns a 0-1 estimate of whether a statement is true — the closest thing to a yes/no question in the set. It is the primitive for gatekeeping: is this on-brand, does this message express urgency, does this draft make a claim we cannot substantiate.
Because each question is independent, adding ten noul checks costs almost nothing in latency compared with adding ten more instructions to a prompt.
- Choice — categorical routing with per-option probabilities
- Score — rubric scoring with a legend
- Noul — truth estimate for a single claim
- Mix freely: one call can carry all three types
How PixaSocial Ai helps
PixaSocial Ai's content workflows already carry the categories and rubrics these primitives need — pillars, channels, personas, guardrails. See System One models for social automation for the applied version.