DecisionMaker class

Performs structured on-device decision making (boolean questions, multiple-choice questions, and scalar scoring) over text or key-value contexts.

Signature:

export declare class DecisionMaker extends TaskRunner

Extends: TaskRunner

Properties

Property Type Description
contextUsage DecisionContextUsage | null Token usage details from the most recent evaluation.
contextWindow number Total token capacity of the underlying model context window.

Methods

Method Modifiers Description
classify(context, options) Convenience method to classify a context into one of the provided option labels.
close() Releases all native WASM resources held by this DecisionMaker instance.
createFromModelBuffer(wasmFileset, modelAssetBuffer) static Initializes the Wasm runtime and creates a new DecisionMaker based on the provided model asset buffer.
createFromModelPath(wasmFileset, modelAssetPath) static Initializes the Wasm runtime and creates a new DecisionMaker based on the path to the model asset.
createFromOptions(wasmFileset, options) static Initializes the Wasm runtime and creates a new DecisionMaker from the provided options.
evaluate(context, questions) Evaluates a heterogeneous list of questions (BooleanQuestion, ChoiceQuestion, ScoreQuestion) against a single shared context KV-cache prefix.
evaluateBatch(items) Evaluates multiple contexts and their associated questions in a batch.
evaluateBoolean(context, question, threshold) Evaluates a yes/no boolean question against the given context.
evaluateBooleanBatch(contexts, question, threshold) Evaluates the same boolean question across a batch of contexts.
evaluateChoice(context, question, options, scoringMode) Evaluates a multiple-choice question with candidate options against the given context.
evaluateChoiceBatch(contexts, question, options, scoringMode) Evaluates the same multiple-choice question across a batch of contexts.
evaluateScore(context, question, scaleMin, scaleMax) Evaluates a scalar rating question over an integer scale [scaleMin, scaleMax].
evaluateScoreBatch(contexts, question, scaleMin, scaleMax) Evaluates the same scalar rating question across a batch of contexts.
measureContextUsage(context, questions) Computes the token count of the given context and questions without running model inference.
prewarm() Pre-warms the underlying inference engine to minimize first-call latency.
setOptions(options) Sets new options for the DecisionMaker instance.
setSchema(schema) Sets or updates the JSON schema / field descriptions used to format structured key-value contexts.

DecisionMaker.contextUsage

Token usage details from the most recent evaluation.

Signature:

get contextUsage(): { contextTokens: number; maxQuestionTokens: number; totalTokens: number; contextWindow: number; } | null;

DecisionMaker.contextWindow

Total token capacity of the underlying model context window.

Signature:

get contextWindow(): number;

DecisionMaker.classify()

Convenience method to classify a context into one of the provided option labels.

Signature:

classify(context: DecisionContext, options: string[]): Promise<ChoiceResult>;

Parameters

Parameter Type Description
context DecisionContext The input text or structured key-value record.
options string[] Candidate category labels.

Returns:

Promise<ChoiceResult>

DecisionMaker.close()

Releases all native WASM resources held by this DecisionMaker instance.

Signature:

close(): void;

Returns:

void

DecisionMaker.createFromModelBuffer()

Initializes the Wasm runtime and creates a new DecisionMaker based on the provided model asset buffer.

Signature:

static createFromModelBuffer(wasmFileset: WasmFileset, modelAssetBuffer: Uint8Array | ReadableStreamDefaultReader): Promise<DecisionMaker>;

Parameters

Parameter Type Description
wasmFileset WasmFileset A configuration object that provides the location of the Wasm binary and its loader.
modelAssetBuffer Uint8Array | ReadableStreamDefaultReader A binary representation of the model or a stream reader.

Returns:

Promise<DecisionMaker>

DecisionMaker.createFromModelPath()

Initializes the Wasm runtime and creates a new DecisionMaker based on the path to the model asset.

Signature:

static createFromModelPath(wasmFileset: WasmFileset, modelAssetPath: TrustedResourceUrl | string): Promise<DecisionMaker>;

Parameters

Parameter Type Description
wasmFileset WasmFileset A configuration object that provides the location of the Wasm binary and its loader.
modelAssetPath TrustedResourceUrl | string The path to the model asset.

Returns:

Promise<DecisionMaker>

DecisionMaker.createFromOptions()

Initializes the Wasm runtime and creates a new DecisionMaker from the provided options.

Signature:

static createFromOptions(wasmFileset: WasmFileset, options: DecisionMakerOptions): Promise<DecisionMaker>;

Parameters

Parameter Type Description
wasmFileset WasmFileset A configuration object that provides the location of the Wasm binary and its loader.
options DecisionMakerOptions Configuration options for the DecisionMaker.

Returns:

Promise<DecisionMaker>

DecisionMaker.evaluate()

Evaluates a heterogeneous list of questions (BooleanQuestion, ChoiceQuestion, ScoreQuestion) against a single shared context KV-cache prefix.

Signature:

evaluate(context: DecisionContext, questions: Array<BooleanQuestion | ChoiceQuestion | ScoreQuestion>): Promise<Array<BooleanResult | ChoiceResult | ScoreResult>>;

Parameters

Parameter Type Description
context DecisionContext The input text or structured key-value context.
questions Array<BooleanQuestion | ChoiceQuestion | ScoreQuestion> Questions to evaluate against the context.

Returns:

Promise<Array<BooleanResult | ChoiceResult | ScoreResult>>

DecisionMaker.evaluateBatch()

Evaluates multiple contexts and their associated questions in a batch.

Signature:

evaluateBatch(items: Array<{ context: DecisionContext; questions: Array<BooleanQuestion | ChoiceQuestion | ScoreQuestion> }>): Promise<Array<Array<BooleanResult | ChoiceResult | ScoreResult>>>;

Parameters

Parameter Type Description
items Array<{ context: DecisionContext; questions: Array<BooleanQuestion | ChoiceQuestion | ScoreQuestion> }> Batch items pairing each context with its questions.

Returns:

Promise<Array<Array<BooleanResult | ChoiceResult | ScoreResult>>>

DecisionMaker.evaluateBoolean()

Evaluates a yes/no boolean question against the given context.

Signature:

evaluateBoolean(context: DecisionContext, question: string | BooleanQuestion, threshold?: number): Promise<BooleanResult>;

Parameters

Parameter Type Description
context DecisionContext The input text or structured context.
question string | BooleanQuestion The boolean question string or BooleanQuestion object.
threshold number Optional probability threshold in [0.0, 1.0] above which answer is true (defaults to 0.5).

Returns:

Promise<BooleanResult>

DecisionMaker.evaluateBooleanBatch()

Evaluates the same boolean question across a batch of contexts.

Signature:

evaluateBooleanBatch(contexts: DecisionContext[], question: string | BooleanQuestion, threshold?: number): Promise<BooleanResult[]>;

Parameters

Parameter Type Description
contexts DecisionContext[] Array of input contexts.
question string | BooleanQuestion The boolean question to evaluate.
threshold number Optional decision threshold in [0.0, 1.0].

Returns:

Promise<BooleanResult[]>

DecisionMaker.evaluateChoice()

Evaluates a multiple-choice question with candidate options against the given context.

Signature:

evaluateChoice(context: DecisionContext, question: string | ChoiceQuestion, options?: string[], scoringMode?: ScoringMode): Promise<ChoiceResult>;

Parameters

Parameter Type Description
context DecisionContext The input text or structured context.
question string | ChoiceQuestion The multiple-choice question string or ChoiceQuestion object.
options string[] Candidate option strings (required when question is a string).
scoringMode ScoringMode Optional scoring mode override.

Returns:

Promise<ChoiceResult>

DecisionMaker.evaluateChoiceBatch()

Evaluates the same multiple-choice question across a batch of contexts.

Signature:

evaluateChoiceBatch(contexts: DecisionContext[], question: string | ChoiceQuestion, options?: string[], scoringMode?: ScoringMode): Promise<ChoiceResult[]>;

Parameters

Parameter Type Description
contexts DecisionContext[] Array of input contexts.
question string | ChoiceQuestion The multiple-choice question.
options string[] Candidate option strings (when question is a string).
scoringMode ScoringMode Optional scoring mode override.

Returns:

Promise<ChoiceResult[]>

DecisionMaker.evaluateScore()

Evaluates a scalar rating question over an integer scale [scaleMin, scaleMax].

Signature:

evaluateScore(context: DecisionContext, question: string | ScoreQuestion, scaleMin?: number, scaleMax?: number): Promise<ScoreResult>;

Parameters

Parameter Type Description
context DecisionContext The input text or structured context.
question string | ScoreQuestion The scoring question string or ScoreQuestion object.
scaleMin number Minimum integer score on the rating scale (defaults to 1).
scaleMax number Maximum integer score on the rating scale (defaults to 5).

Returns:

Promise<ScoreResult>

DecisionMaker.evaluateScoreBatch()

Evaluates the same scalar rating question across a batch of contexts.

Signature:

evaluateScoreBatch(contexts: DecisionContext[], question: string | ScoreQuestion, scaleMin?: number, scaleMax?: number): Promise<ScoreResult[]>;

Parameters

Parameter Type Description
contexts DecisionContext[] Array of input contexts.
question string | ScoreQuestion The scoring question.
scaleMin number Minimum integer score (defaults to 1).
scaleMax number Maximum integer score (defaults to 5).

Returns:

Promise<ScoreResult[]>

DecisionMaker.measureContextUsage()

Computes the token count of the given context and questions without running model inference.

Signature:

measureContextUsage(context: DecisionContext, questions?: Array<BooleanQuestion | ChoiceQuestion | ScoreQuestion>): Promise<{ contextTokens: number; maxQuestionTokens: number; totalTokens: number; contextWindow: number; }>;

Parameters

Parameter Type Description
context DecisionContext The input context to measure.
questions Array<BooleanQuestion | ChoiceQuestion | ScoreQuestion> Optional questions to include in the token count.

Returns:

Promise<{ contextTokens: number; maxQuestionTokens: number; totalTokens: number; contextWindow: number; }>

DecisionMaker.prewarm()

Pre-warms the underlying inference engine to minimize first-call latency.

Signature:

prewarm(): Promise<void>;

Returns:

Promise<void>

DecisionMaker.setOptions()

Sets new options for the DecisionMaker instance.

Signature:

setOptions(options: DecisionMakerOptions): Promise<void>;

Parameters

Parameter Type Description
options DecisionMakerOptions New options to apply.

Returns:

Promise<void>

DecisionMaker.setSchema()

Sets or updates the JSON schema / field descriptions used to format structured key-value contexts.

Signature:

setSchema(schema: Record<string, string> | string): Promise<void>;

Parameters

Parameter Type Description
schema Record<string, string> | string Mapping of field names to human-readable descriptions, or a serialized JSON schema string.

Returns:

Promise<void>