/**
 * TotalReclaw LSH Hasher (WASM-backed)
 *
 * Thin wrapper over `WasmLshHasher` from `@totalreclaw/core`. Same class
 * interface as the previous pure-TS implementation so callers don't need
 * to change.
 *
 * Default parameters:
 *   - 32 bits per table (balanced discrimination vs. recall)
 *   - 20 tables (moderate table count for good coverage)
 *
 * Matches mcp/src/subgraph/lsh.ts exactly.
 */
/**
 * Random Hyperplane LSH hasher.
 *
 * All state is deterministic from the seed -- no randomness at hash time.
 * Construct once per session; call `hash()` for every store/search operation.
 */
export declare class LSHHasher {
    private inner;
    /** Embedding dimensionality (cached for error messages). */
    private readonly dims;
    /**
     * Create a new LSH hasher.
     *
     * @param seed    - 32-byte seed from `deriveLshSeed()` in seed.ts.
     * @param dims    - Embedding dimensionality (e.g. 640 for Harrier-OSS-v1-270M).
     * @param nTables - Number of independent hash tables (default 20).
     * @param nBits   - Number of bits per table (default 32).
     */
    constructor(seed: Uint8Array, dims: number, nTables?: number, nBits?: number);
    /**
     * Hash an embedding vector to an array of blind-hashed bucket IDs.
     *
     * For each table:
     *   1. Compute the N-bit signature (sign of dot product with each hyperplane).
     *   2. Build the bucket string: `lsh_t{tableIndex}_{binarySignature}`.
     *   3. SHA-256 the bucket string to produce a blind hash (hex).
     *
     * @param embedding - The embedding vector (must have `dims` elements).
     * @returns Array of `nTables` hex strings (one blind hash per table).
     */
    hash(embedding: number[]): string[];
    /** Number of hash tables. */
    get tables(): number;
    /** Number of bits per table. */
    get bits(): number;
    /** Embedding dimensionality. */
    get dimensions(): number;
}
/**
 * Compute the Hamming distance between two binary signature strings.
 *
 * @param bucket1 - First bucket ID (binary string)
 * @param bucket2 - Second bucket ID (binary string)
 * @returns Number of differing bits
 */
export declare function hammingDistance(bucket1: string, bucket2: string): number;
/**
 * Estimate similarity from Hamming distance.
 *
 * For random hyperplane LSH, the expected Hamming distance
 * is proportional to the angle between vectors.
 */
export declare function estimateSimilarity(hammingDist: number, nBits: number): number;
//# sourceMappingURL=hyperplane.d.ts.map