/**
 * Tropical (min-plus) sparse matrix–vector multiply on CSR graphs.
 *
 * Matches the semiring used in `@holoscript/snn-webgpu` / `TropicalShortestPaths`
 * (`INF = 1e30`). One SpMV step computes
 *   y_i = min_j (A_ij + x_j)
 * over stored entries only.
 *
 * Used for CPU reference checks, graph benchmarks (ER / scale-free / layered),
 * and comparison against the dense min-plus path.
 */
/** Same sentinel as `TropicalShortestPaths` / WGSL kernels. */
export declare const TROPICAL_INF = 1e+30;
/** CSR adjacency (square n×n), row-major compressed sparse rows. */
export interface TropicalCsrMatrix {
    readonly n: number;
    readonly rowPtr: Uint32Array;
    readonly colIdx: Uint32Array;
    readonly values: Float32Array;
}
/** Deterministic PRNG (Mulberry32) for reproducible graph builds. */
export declare function mulberry32(seed: number): () => number;
/**
 * Min-plus SpMV: y_i = min_k (values[k] + x[colIdx[k]]) for entries in row i.
 * Rows with no entries get y_i = TROPICAL_INF.
 */
export declare function tropicalMinPlusSpmv(csr: TropicalCsrMatrix, x: Float32Array, y: Float32Array): void;
/** Dense min-plus SpMV (reference): y_i = min_j (A_ij + x_j). */
export declare function tropicalMinPlusSpmvDense(n: number, aRowMajor: Float32Array, x: Float32Array, y: Float32Array): void;
/** Build CSR from dense matrix, storing only finite entries strictly below INF. */
export declare function csrFromDense(n: number, dense: Float32Array): TropicalCsrMatrix;
/** Directed Erdős–Rényi G(n,p): edge i→j with probability p, weight in (0,1]. */
export declare function erdosRenyiCsr(n: number, p: number, rng: () => number): TropicalCsrMatrix;
/**
 * Barabási–Albert–style directed graph: each new vertex adds up to `m` outgoing
 * edges to strictly older vertices; targets are chosen with probability
 * proportional to (indegree + 1).
 */
export declare function barabasiAlbertCsr(n: number, m: number, rng: () => number): TropicalCsrMatrix;
/** Layered feedforward with random skip connections (neural-graph toy). */
export declare function layeredNeuralCsr(n: number, layers: number, skipProbability: number, rng: () => number): TropicalCsrMatrix;
/** Max absolute difference between two vectors (skipping mutual INF). */
export declare function maxAbsDiff(a: Float32Array, b: Float32Array): number;