/**
 * Console reporter for the planRun-based pipeline.
 *
 * Owns its own progress UI (spinner + status table) and serialized
 * stdout writes. Internally aggregates per-stimulus trial blocks for
 * multi-trial runs so the user sees one tidy block per stimulus rather
 * than interleaved partial results from concurrent workers.
 *
 * Layout per event:
 *   onRunStart           → header lines, start spinner with phase table
 *   onTrialStart         → mark slot `queued → starting`
 *   onTrialPhase         → update slot phase
 *   onTrialResult        → for single-trial: flush per-stimulus block
 *                          for multi-trial: append to stimulus bucket;
 *                          when bucket has totalTrials, flush block
 *   onEvalComplete       → per-eval verdict line
 *   onRunComplete        → stop spinner, cross-model summary,
 *                          artifacts list
 */
import type { PlanReporter, PlanRunStartContext, RunArtifacts } from "./plan-reporter.js";
import type { EvalSummary, RunSummary, ScopedDiagnostic, TrialPhase, TrialPhaseDetail, TrialResult, TrialWorkItem } from "../pipeline/plan.js";
import { sanitizeForDisplay } from "./sanitize.js";
export { sanitizeForDisplay };
export interface EvalConsoleReporterOptions {
    verbose?: boolean;
    /** Where rendered blocks land. Defaults to `process.stdout`. */
    stream?: NodeJS.WritableStream;
    /** Where the spinner renders. Defaults to `process.stderr`. */
    progressStream?: NodeJS.WriteStream;
    /** Disable spinner explicitly (useful for tests / non-TTY). */
    disableSpinner?: boolean;
}
export declare class EvalConsoleReporter implements PlanReporter {
    private readonly verbose;
    private readonly stream;
    private readonly mutex;
    private readonly spinner;
    private readonly slotsById;
    private nextSlotIndex;
    private completed;
    private totalItems;
    private workers;
    private readonly buckets;
    /** Effective threshold per (eval × variant × model), captured at
     *  onRunStart. Drives the threshold-adjusted per-trial pass in
     *  multi-trial blocks so glyphs/counts agree with pass@k and the
     *  eval verdict. undefined when scoring isn't applied. */
    private readonly thresholdByEval;
    /** Variant names present in the plan, captured at onRunStart so
     *  per-trial headers can include the variant only when more than one
     *  is in play (single-variant runs would otherwise show every header
     *  with a redundant "main" label). */
    private readonly variantNames;
    private summaryHeadingShown;
    constructor(options?: EvalConsoleReporterOptions);
    onRunStart(ctx: PlanRunStartContext): Promise<void>;
    onTrialStart(item: TrialWorkItem): Promise<void>;
    onTrialPhase(itemId: string, phase: TrialPhase, detail?: TrialPhaseDetail): Promise<void>;
    onTrialResult(event: {
        item: TrialWorkItem;
        result: TrialResult;
    }): Promise<void>;
    onDiagnostics(diagnostics: ScopedDiagnostic[]): Promise<void>;
    onEvalComplete(summary: EvalSummary): Promise<void>;
    /**
     * Stop the live progress spinner without rendering the run summary. Safe to
     * call multiple times and when no spinner is active; used on the error path so
     * the terminal isn't left spinning when a run aborts before completion.
     */
    stopProgress(): void;
    onRunComplete(summary: RunSummary, artifacts: RunArtifacts): Promise<void>;
    /**
     * Format the `━━━ name · stim [variant] [model] ━━━` header used for
     * both single-trial and multi-trial blocks. The variant tag only
     * appears when the plan has more than one variant — single-variant
     * runs (which is every `vally eval` invocation) would otherwise
     * decorate every header with a redundant default name.
     */
    private formatTrialHeader;
    private renderSingleTrialBlock;
    private renderMultiTrialBlock;
    private renderMetricsBlock;
    private renderGradedBlock;
    /**
     * Render one `icon label  evidence` row per grader, plus indented rows for
     * the failing sub-checks of a failing grader. Graders that share a name get
     * a `#n` suffix so each row stays distinguishable.
     *
     * Grader names and evidence originate in user-controlled YAML and
     * model-generated text, so every interpolated string is sanitized.
     */
    private renderGraderRows;
    private renderAgentOutputBlock;
    private flushBlock;
    private write;
    private spinnerText;
    private refreshSpinnerStatus;
    private refreshSpinnerTable;
}
//# sourceMappingURL=eval-console.d.ts.map