CouCouGen.js

Ivan Petrov

CouCouGen.js is a lightweight, zero-dependency deterministic sequence generator for JavaScript. Based on three coupled 32-bit counters with nonlinear feedback, it achieves a period of up to 2^96 (~7.9 × 10^28 numbers) — theoretical, not proven. NOT a statistically proven PRNG. Independent Dieharder testing (100M numbers) reveals failures in bit-level tests: rank_32x32, bitstream, DNA, count_1s, birthdays. Passes basic uniformity tests (operm5, spheres, runs, craps). Key features: - State size: 96 bits (3 × 32-bit counters) - Speed: ~5.6 ms per 100,000 numbers - Size: less than 2 KB uncompressed - No external dependencies - MIT License Basic statistical properties (tested on 100,000 numbers): - Mean: 0.500036 (ideal: 0.5) - Standard deviation: 0.288673 (ideal: ~0.288675) - Chi-square (20 bins): 19.02 (ideal: ~19.0) - Autocorrelation (lag 1): 0.00245 Package includes: - coucougen.js - main library (v1.2.0) - README.md / README-ru.md - full documentation - PAPER.md - bilingual paper with mathematical analysis (Dieharder results, open questions) - CHANGELOG.md - version history - package.json - npm metadata - Statistical analysis tools Suitable for: games, procedural generation (full 32-bit integers), visualizations, prototyping, testing, education. NOT suitable for: bit-level operations (value & 1), Monte Carlo simulations, matrix operations, cryptography, security, gambling, critical infrastructure. Author: Ivan Petrov License: MIT Year: 2026 Version: 1.2.0