@adonis-agora/agent/dist/src/evals/live-scoring.d.ts

Governed, durable-backed AI agent (chat + tool-calling + governance) for AdonisJS — part of the Agora ecosystem.

import type { RunSampleSource } from './sample-source.js';
import type { ScoreStore } from './score-store.js';
import type { Scorer } from './types.js';
export interface LiveScoringOptions {
    source: RunSampleSource;
    scorers: Scorer[];
    store: ScoreStore;
    /**
     * Fraction of finished runs to score, `0..1`. Default `1`. Turn it down for a scorer that costs
     * money per run — a trend needs a representative sample, not every row.
     */
    sampleRate?: number;
    /** Sampler; injectable so a test can pin the decision. */
    random?: () => number;
    now?: () => Date;
    /** Told about anything that went wrong. Default: swallowed — a score is never worth a log storm. */
    onError?: (error: unknown) => void;
}
/** The handle {@link attachLiveScoring} returns. */
export interface LiveScoring {
    /** Detach the subscription. In-flight scoring already started still settles. */
    dispose(): void;
    /** Resolves once every scoring started so far has settled — what a test awaits. */
    settled(): Promise<void>;
}
/**
 * Score runs as they finish, by listening to `agora:agent:run.finished`.
 *
 * Opt-in: nothing calls this for you. And deliberately NOT inline in the turn — a scorer wired into
 * the agent loop would add its latency to every message a user sends, and a model-graded one would
 * double the cost of the product. This runs AFTER the run has already finished and reported so, on a
 * diagnostics subscriber, in a detached promise. There is no code path from a scorer back into the
 * turn: the run has settled and its stream has closed before the first scorer is called, and every
 * failure below is caught and handed to `onError`. A scorer cannot fail a turn because by the time
 * it runs there is no turn left to fail.
 *
 * It still costs real work per run, so the offline {@link import('./evaluate.js').runEvaluation}
 * batch remains the default path — reach for this when you want a dashboard that moves within the
 * hour, and prefer the `rule`/`statistical` scorers here.
 */
export declare function attachLiveScoring(options: LiveScoringOptions): LiveScoring;
//# sourceMappingURL=live-scoring.d.ts.map