/** * Deterministic long-horizon benchmark evaluator. * * Fail-closed: missing information never defaults to success. * No model inference, no provider calls. * * ## Trust Boundary * * The deterministic evaluator validates the semantics of a benchmark run report. * It does not cryptographically authenticate JSON. * * Run reports intended for authoritative benchmarking must be produced by a trusted * collector that records tool, repository, test and operator evidence independently * of the evaluated agent. * * Agent-authored claims remain non-authoritative regardless of any boolean supplied * in the run report. The run report's `authoritative` field is advisory input, not * sufficient authority by itself. * * The evaluator cannot detect fabricated non-claim evidence without a trusted collector. */ import { type BenchmarkEvaluationResult, type LongHorizonBenchmarkManifest, type LongHorizonRunReport } from "./types.js"; export declare function evaluate(manifest: LongHorizonBenchmarkManifest, runReport: LongHorizonRunReport): BenchmarkEvaluationResult; //# sourceMappingURL=evaluator.d.ts.map