/**
 * Grading — run graders against trajectories, optionally compare two runs.
 *
 * Core function:
 *   gradeTrajectory(trajectory, graders)   → grade a single trajectory
 *   compareRuns(baseline, skilled)          → grade both, compute deltas
 *
 * This is pure evaluation — no agent execution happens here.
 */
import type { Trajectory } from "../trajectory/types.js";
import { type GraderResult } from "../graders/types.js";
import { type Stimulus, type StimulusGraderConfig } from "../eval/types.js";
import type { GraderRegistry } from "../graders/registry.js";
import type { LlmClient } from "../graders/llm/types.js";
import type { ReasoningEffort } from "../graders/llm/types.js";
import type { JudgeProviderSpec } from "../provider/copilot-provider.js";
export interface StimulusGradeResult extends GraderResult {
    stimulusName: string;
    trajectoryId: string;
    timestamp: Date;
}
export interface ComparisonResult {
    baseline: GraderResult;
    skilled: GraderResult;
    deltas: MetricDeltas;
}
export interface MetricDeltas {
    tokensDelta: number;
    turnsDelta: number;
    toolCallsDelta: number;
    wallTimeMsDelta: number;
    errorsDelta: number;
    skillActivated: boolean;
}
export interface GradeTrajectoryOptions {
    registry?: GraderRegistry;
    /**
     * Authoritative stimulus from the eval spec. For turn-scoped diff grading,
     * this must preserve the original `turns` structure and must not be rebuilt
     * from a persisted trajectory's synthesized `prompt`.
     */
    stimulus?: Stimulus;
    /** Default model for LLM graders (from eval config or CLI flag). Per-grader config.model takes priority. */
    judgeModel?: string;
    /**
     * Reasoning effort for the judge model. Applied only when the eval-level
     * judge model is used (i.e. no per-grader `config.model` override).
     */
    judgeReasoningEffort?: ReasoningEffort;
    /**
     * BYOK provider for the LLM judge (from `defaults.judge_provider`). Injected
     * into every LLM grader so the judge session targets a custom endpoint.
     */
    judgeProvider?: JudgeProviderSpec;
    /** Grader weights from `scoring.weights` — keyed by grader type. Must be a normalized distribution (values sum to 1). When provided, the aggregate score is the weighted sum; types absent from the map receive weight 0 and are excluded from the score. */
    weights?: Record<string, number>;
    /** Base directory for resolving relative sidecar diff paths (trajectory.diffPath). Pass the --run-dir value when re-grading from a saved run output directory. */
    diffBaseDir?: string;
    /** Reference-solution diff passed to every grader as `GraderInput.goldenPatch`. */
    goldenPatch?: string;
    /** Optional callback fired before each grader runs. */
    onGraderStart?: (info: {
        graderIndex: number;
        totalGraders: number;
        graderName: string;
        graderType: string;
    }) => void | Promise<void>;
    /** Optional callback fired after each grader completes. */
    onGraderComplete?: (info: {
        graderIndex: number;
        totalGraders: number;
        graderName: string;
        passed: boolean | null;
        evidence: string;
        durationMs: number;
    }) => void | Promise<void>;
}
export declare function createDefaultGraderRegistry(llmClient?: LlmClient): GraderRegistry;
/**
 * Register built-in LLM graders into an existing registry.
 *
 * Use this to upgrade a registry with LLM capabilities without
 * recreating it (which would drop any plugin registrations).
 */
export declare function registerLlmGraders(registry: GraderRegistry, llmClient: LlmClient): void;
/**
 * Check whether any grader configs require an LLM client.
 *
 * Checks the registry first (catches plugin/external graders via the
 * behavior contract), then falls back to built-in LLM grader names for
 * cases where the LLM graders aren't yet registered (no LlmClient provided).
 */
export declare function needsLlmClient(graderConfigs: StimulusGraderConfig[], registry?: GraderRegistry): boolean;
/**
 * Check whether any grader configs require filesystem access to the workspace.
 *
 * Checks the registry first (catches plugin/external graders via the
 * behavior contract), then falls back to built-in workspace grader names
 * for cases where graders aren't yet registered.
 */
export declare function needsWorkspace(graderConfigs: StimulusGraderConfig[], registry?: GraderRegistry): boolean;
export declare function gradeTrajectory(trajectory: Trajectory, graderConfigs: StimulusGraderConfig[], options?: GradeTrajectoryOptions): Promise<GraderResult>;
export declare function compareRuns(baseline: Trajectory, skilled: Trajectory, graderConfigs: StimulusGraderConfig[], registry?: GraderRegistry, weights?: Record<string, number>, stimulus?: Stimulus): Promise<ComparisonResult>;
export declare function mapGraderConfig(graderConfig: StimulusGraderConfig): Record<string, unknown>;
//# sourceMappingURL=grading.d.ts.map