import type { Claim, LedgerRecord } from '../core/record.js';
import type { ExecFn } from '../util/exec.js';
/**
 * "The agent says the tests passed" is the single most consequential claim in
 * a coding run, so it gets first-class verification: re-run the exact command
 * and compare. This is the replay half of "logs are claims, replays are
 * proofs" — deterministic, no model in the loop.
 */
export type TestClaimStatus = 'verified' | 'now_failing' | 'mismatch' | 'skipped' | 'error';
export interface TestClaimVerification {
    readonly seq: number;
    readonly command: string;
    readonly claimedOutcome: 'ok' | 'failed' | 'unknown';
    readonly status: TestClaimStatus;
    readonly detail: string;
}
export interface VerifyTestsOptions {
    readonly rerun: boolean;
    readonly cwd: string;
    readonly timeoutMs?: number;
    readonly exec?: ExecFn;
    /** Re-run only the last N test claims (the final state is what you ship). */
    readonly lastN?: number;
}
export declare function verifyTestClaims(records: readonly LedgerRecord[], options: VerifyTestsOptions): readonly TestClaimVerification[];
export declare function collectTestClaims(records: readonly LedgerRecord[]): ReadonlyArray<{
    seq: number;
    claim: Extract<Claim, {
        type: 'test_run';
    }>;
}>;