/**
 * Budgeting — the deterministic estimation model (pure). Converts a story's
 * COUNTABLE drivers (spec structure) + the LLM's RELATIVE `effort` block into a
 * 3-point, dual-currency estimate (human manhours + agent tokens), then rolls a
 * backlog up to portfolio percentiles via Monte-Carlo simulation.
 *
 * Boundary: the LLM sizes relative + names qualitative drivers (it's unreliable at
 * absolute hours); THIS model owns the absolute conversion via calibrated
 * coefficients. The seed coefficients below are illustrative defaults — the real
 * ones are LEARNED from actuals (reference-class forecasting / the SizeCostTable).
 * See dash docs/BUDGETING-AND-ROADMAP.md.
 */
import type { Spec, EffortSize } from "@slowcook-ai/core";
/** Countable complexity drivers extracted from the spec structure (the inside view). */
export interface StoryDrivers {
    entities: number;
    fields: number;
    relations: number;
    endpoints: number;
    surfaces: number;
    /** Σ declared states across surfaces (≥1 per surface) — the EPSS matrix size. */
    epssCells: number;
    scenarios: number;
    invariants: number;
    fidelityModes: number;
    openQuestions: number;
    personas: number;
}
/** A 3-point estimate (optimistic / most-likely / pessimistic) in one unit. */
export interface ThreePoint {
    o: number;
    m: number;
    p: number;
}
/** A story's estimate across both currencies, 3-point each. */
export interface StoryEstimate {
    storyId: string;
    title: string;
    epic: string | null;
    /** human manhours (design + qa + review). */
    hours: ThreePoint;
    /** agent tokens (build + test). */
    tokens: ThreePoint;
    /** total cost in cents (labor + compute). */
    costCents: ThreePoint;
    risk: "low" | "medium" | "high";
    confidence: number;
    drivers: StoryDrivers;
    qualitativeDrivers: string[];
}
/** Calibration — seed coefficients (illustrative; calibrate from actuals). */
export interface Calibration {
    /** relative weight per t-shirt size. */
    weight: Record<EffortSize, number>;
    /** human hours per design/qa weight unit. */
    designHoursPerWeight: number;
    qaHoursPerWeight: number;
    /** agent k-tokens per build weight unit. */
    buildKTokPerWeight: number;
    /** structural coefficients (the inside view). */
    structural: {
        designBase: number;
        dSurf: number;
        dMode: number;
        buildBaseK: number;
        bEntity: number;
        bField: number;
        bApi: number;
        bSurf: number;
        qaBase: number;
        qCell: number;
        qInv: number;
        testBaseK: number;
        tScenario: number;
        reviewBase: number;
        rSurf: number;
        rInv: number;
    };
    /** multiplier per qualitative driver (default 1.0 when absent). */
    driverMult: Record<string, number>;
    /** added band per risk level. */
    riskAdd: Record<"low" | "medium" | "high", number>;
    baseBand: number;
    /** $/hour (cents) blended labor rate. */
    roleRateCents: number;
    /** $/million-tokens (cents) blended compute rate. */
    tokenRateCentsPerM: number;
}
export declare const SEED_CALIBRATION: Calibration;
/** Count the structural drivers from a spec (the inside view). */
export declare function extractDrivers(spec: Spec): StoryDrivers;
/** Estimate one story: countable drivers ⊕ the LLM effort block → 3-point dual-currency. */
export declare function estimateStory(spec: Spec, calib?: Calibration): StoryEstimate;
/** Sample a triangular(o,m,p) distribution given u∈[0,1). */
export declare function sampleTriangular(tp: ThreePoint, u: number): number;
export interface Percentiles {
    p50: number;
    p85: number;
    p95: number;
}
export interface PortfolioForecast {
    hours: Percentiles;
    tokens: Percentiles;
    costCents: Percentiles;
    /** the deterministic Σ of most-likely values (the v1 method) for comparison. */
    deterministic: {
        hours: number;
        tokens: number;
        costCents: number;
    };
    iterations: number;
    stories: number;
}
/** Monte-Carlo: per iteration draw a systemic factor, then sample each story's
 *  3-point per currency, sum, report percentiles. */
export declare function monteCarloPortfolio(estimates: StoryEstimate[], iterations?: number, seed?: number): PortfolioForecast;
//# sourceMappingURL=model.d.ts.map