The three question kinds
Every question is one of three kinds, chosen by itstype. Each carries free-text
instructions and, optionally, criteria that pin down what the outcomes mean.
noul— a binary judgment. The classifier returns the probability, from 0 to 1, that the answer is yes. Optionalcriteriadescribe what a true and a false answer mean. For example, over an inbound event: “Is this event urgent?” →0.82.choice— a categorical judgment.criteriais a fixed set of labels, each mapped to a description; the classifier picks one label and reports a probability for every label. For example, over an account: “Which status fits?” with labelsactive,dormant,closed→activeat0.7. The labels are the question’s allowed set, and the platform enforces it (see Answers stay inside what the request declares).score— an ordinal judgment.criteriais an ordered rubric of two or more levels; the classifier returns the expected value over that rubric plus the per-level distribution. For example, over an alert: “Rate the severity” over["low", "medium", "high", "critical"]→ an expected score of2.4.
A request and its response
A request pairs the state with a named mapping of questions — the name is the author’s key for that question:type discriminator:
noul answer carries just its noul probability; a choice answer carries the
chosen choice, the per-label probabilities, and a confidence; a score answer
carries the expected score, the level legend and probabilities keyed by level
index, and a confidence. Either token count may be unreported.
Answers stay inside what the request declares
A request declares a set of answers it will accept, and the platform checks every response against it, whatever the provider:- The questions. The answers are keyed by exactly the request’s question names, one answer per question, and each answer is of its question’s kind.
- A
choicequestion’s labels. The labels are the keys of itscriteria. The chosenchoicemust be one of them, and theprobabilitiesmust be keyed by exactly the labels, one probability per label. - A
scorequestion’s levels. The levels are0up to one less than the number of rubric entries. Thelegendand theprobabilitiesmust be keyed by exactly the levels, and thescoremust lie between the first and the last level.
ClassifierAnswerError naming the question, what the provider gave, and what the request
declared. Its error kind is upstream_error, because the request was valid and the
provider broke the contract. A caller, a tool, or any other run can therefore read every answer
as its own question declared it, without checking it.
Configuring a classifier
A deployment picks its classifier provider with theLLM_PROVIDER_CLASSIFIER
selector — the third model-provider selector beside LLM_PROVIDER_LLM and
LLM_PROVIDER_EMBEDDING — which defaults to typesafe. The provider’s connection
details ride the CLASSIFIER_* namespace: CLASSIFIER_MODEL (default jev-latest),
CLASSIFIER_BASE_URL, CLASSIFIER_API_KEY, and CLASSIFIER_TIMEOUT. The full field
table is in the settings reference, and the
config and secrets page covers the
provider selectors.
The classifier provider ships as an optional dependency. The classify tool pulls it
in through the classifier extra — install it with pip install 'tai42-toolbox[classifier]', which brings the kit’s typesafe provider along.
Where classifiers are used
- The
classifytool. The standard toolbox exposes classification as a tool any caller can run — see theclassifywalk-through. - The Classifier card in the Studio. A Babelfish Flows flow can judge its input with a Classifier card, whose answers feed downstream conditions.
- The Router card’s classifier backing. A Router can be decided by a classifier model instead of an LLM agent, asking one yes/no question per kind and routing on the probability.
- The kit factory. In Python,
get_classifier(andget_classifier_async) build a cached, provider-keyed runnable that takes a request and returns a response — see the Python SDK reference.
classify, the
Babelfish Flows page for judging inside a flow,
and the settings reference for every
CLASSIFIER_* field.
