/**
 * CLIFF / TIPPING-POINT detector — where does a tiny change make the result fall off a cliff? Most analyses
 * assume the response is smooth. Real processes aren't: a catalyst poisons past a temperature, a model
 * collapses past a learning rate, a material shatters past a load, a yield craters past a pH. These cliffs
 * are where disasters live — and a "best setting" sitting right on the edge of one is a setting that works
 * brilliantly today and fails the morning the room is 1° warmer.
 *
 * CLIFF scans your measurements for NEIGHBOURS that are close in settings but far apart in result — the
 * fingerprint of a cliff — and reports where they are, how big the drop is, and which knob you crossed.
 * Crucially, it flags when your OPTIMUM sits on a cliff edge: a loud warning to step back to a safer,
 * flatter setting even if it scores a hair lower.
 *
 * Honest by construction (DIAKRISIS): a cliff is detected only where two genuinely-near experiments disagree
 * far more than the typical local change AND the drop is a real fraction of your result range — measured
 * from your data, not assumed. It can't see a cliff hiding between samples you never took near; it reports
 * the ones your data actually straddles, and abstains when data is too thin.
 */
import { type Space } from "./space.js";
import { type Observation, type Goal } from "./engine.js";
export interface Cliff {
    at: Record<string, number>;
    drop: number;
    steepness: number;
    variable: string;
}
export interface CliffReport {
    cliffs: Cliff[];
    optimumOnCliff: boolean;
    note: string;
}
/** Find settings where a small change causes a big drop in result (cliffs / tipping points). */
export declare function analyzeCliffs(obs: ReadonlyArray<Observation>, space: Space, _goal?: Goal): CliffReport;
export declare function cliffGauntlet(): {
    score: 0 | 100;
    checks: Array<{
        name: string;
        pass: boolean;
        detail: string;
    }>;
};
//# sourceMappingURL=cliff.d.ts.map