Embabel Agent Documentation/com.embabel.common.ai.decision.support/StubDecisionService

StubDecisionService

@ApiStatus.Experimental

sealed class StubDecisionService : DecisionService, PropositionAssessment, RatingAssessment

A decision service for tests that returns scripted outcomes and records each call.

Outcomes for ask, rate and question assess calls are scripted by question name, and outcomes for assess calls with a PropositionRequest by proposition text. classify returns the choice scripted for the request's question name, and otherwise the classification scripted for its set of category ids. A call with no scripted outcome throws IllegalStateException.

By default the stub answers a whole request in one question-set call. Question-set answers pass through the same question validation as provider answers, so a scripted choice outside its question's options, or rating evidence outside its levels, throws IllegalStateException from ask. A stub built after Builder.perQuestion answers each question on its own: a choice question through classify, and the other kinds through their per-question hooks. The stub accepts every question kind unless other capabilities are set.

val stub = StubDecisionService.builder("triage-stub")
    .proposition("is_urgent", PropositionResult.Answered(true, provenance))
    .build()

Types

Builder

@ApiStatus.Experimental

class Builder

Collects scripted outcomes for a StubDecisionService. Get one from StubDecisionService.builder. A later outcome for the same key replaces the earlier one.

Companion

object Companion

Creates stub decision services.

Properties

name

open override val name: String

Name of the LLM, such as "gpt-3.5-turbo"

provider

open override val provider: String

Name of the provider, such as "OpenAI"

type

open override val type: ModelType

Family remains DECISION even when this service is used only for classification.

Functions

ask

open fun ask(request: DecisionRequest): DecisionResponse

Answers a request.

open fun ask(input: String, spec: DecisionSpec): DecisionResponse

Answers the questions of a spec against the given input. It behaves as ask with a request.

assess

open override fun assess(request: PropositionRequest): PropositionResult

Assess the proposition; a false answer is successful evidence, not an operational failure.

open override fun assess(input: String, question: PropositionQuestionSpec): PropositionResult

Answers one proposition question in its own provider call.

calls

fun calls(): List<String>

Returns the names of the operations called so far, in call order: askQuestionSet, assess, classify and rate. A question assess call is recorded as assess.

capabilities

open override fun capabilities(): DecisionCapabilities

Returns what this service can accept.

classify

open fun classify(input: String, spec: ClassificationSpec): ClassificationResult

Classifies text with a spec. Same as calling classify with ClassificationRequest.of(input, spec).

open override fun classify(request: ClassificationRequest): ClassificationResult

Classify text using only the request's category domain, returning evidence rather than routing policy.

infoString

open fun infoString(verbose: Boolean?, indent: Int): String

metadata

open override fun metadata(): DecisionServiceMetadata

Pure DECISION metadata snapshot; never serialize a live service in place of this value.

rate

open override fun rate(input: String, question: RatingQuestionSpec): RatingResult

Rates the input against one rating question in its own provider call. The caller validates the outcome against the question's levels.

toString

open override fun toString(): String