/**
 * [WHO]: Versioned real-task benchmark snapshot, policy, metric, check, and promotion-report contracts
 * [FROM]: No runtime dependencies
 * [TO]: Consumed by evolution benchmark validation, comparison, CLI, gate, and store
 * [HERE]: extensions/optional/evolution/benchmark-types.ts - private evolution evidence protocol
 */
export type EvolutionBenchmarkRole = "baseline" | "candidate";
export type EvolutionBenchmarkSplit = "train" | "validation" | "heldout";
export interface EvolutionBenchmarkRunV1 {
    taskId: string;
    repetition: number;
    split: EvolutionBenchmarkSplit;
    slices: string[];
    diagnostics?: string[];
    success: boolean;
    score: number;
    costUsd: number;
    latencyMs: number;
    policyViolations: number;
    replayDivergences: number;
    unpairedToolCalls: number;
}
export interface EvolutionBenchmarkSnapshotV1 {
    schemaVersion: 1;
    kind: "catui-evolution-benchmark-snapshot";
    role: EvolutionBenchmarkRole;
    candidateId?: string;
    candidateContentHash?: string;
    createdAt: string;
    corpus: {
        id: string;
        version: string;
        digest: string;
    };
    harness: {
        revisionId: string;
        commitSha: string;
    };
    execution: {
        model: string;
        modelVersion: string;
        temperature: number;
        maxTokens: number;
        timeoutMs: number;
        budgetUsd: number;
    };
    runs: EvolutionBenchmarkRunV1[];
}
export interface EvolutionBenchmarkPolicyV1 {
    schemaVersion: 1;
    minimumRepetitions: number;
    minimumPassRateGain: number;
    confidenceLevel: number;
    bootstrapIterations: number;
    maximumSliceRegression: number;
    maximumCostPerSuccessRegression: number;
    maximumP95LatencyRegression: number;
}
export type EvolutionBenchmarkCheckId = "minimum_pass_rate_gain" | "positive_confidence_lower_bound" | "maximum_slice_regression" | "zero_policy_violations" | "zero_replay_divergences" | "zero_unpaired_tool_calls" | "maximum_cost_per_success_regression" | "maximum_p95_latency_regression";
export interface EvolutionBenchmarkCheckV1 {
    id: EvolutionBenchmarkCheckId;
    passed: boolean;
    actual: number;
    threshold: number;
    summary: string;
}
export interface EvolutionBenchmarkMetricsV1 {
    heldoutTasks: number;
    pairedRuns: number;
    baselinePassRate: number;
    candidatePassRate: number;
    passRateGain: number;
    confidenceLowerBound: number;
    sliceDeltas: Record<string, number>;
    maximumSliceRegression: number;
    policyViolations: number;
    replayDivergences: number;
    unpairedToolCalls: number;
    baselineCostPerSuccess: number;
    candidateCostPerSuccess: number;
    costPerSuccessRegression: number;
    baselineP95LatencyMs: number;
    candidateP95LatencyMs: number;
    p95LatencyRegression: number;
}
export interface EvolutionBenchmarkPromotionReportV1 {
    schemaVersion: 1;
    kind: "catui-evolution-benchmark-promotion-report";
    candidateId: string;
    candidateContentHash: string;
    /**
     * Revision the candidate was created against, or null when none was active. Binding the
     * baseline here means a report gathered against one active state cannot validate a candidate
     * that would replace a different one.
     */
    baselineRevisionId: string | null;
    checkedAt: string;
    corpus: EvolutionBenchmarkSnapshotV1["corpus"];
    baselineSnapshotHash: string;
    candidateSnapshotHash: string;
    policy: EvolutionBenchmarkPolicyV1;
    metrics: EvolutionBenchmarkMetricsV1;
    checks: EvolutionBenchmarkCheckV1[];
    passed: boolean;
    contentHash: string;
}
export interface ValidatedEvolutionBenchmarkPair {
    baseline: EvolutionBenchmarkSnapshotV1;
    candidate: EvolutionBenchmarkSnapshotV1;
    baselineHeldout: EvolutionBenchmarkRunV1[];
    candidateHeldout: EvolutionBenchmarkRunV1[];
}