A Class-based Random Number Generator with Entropy Strength Classification for NIST-Compliant Cryptographic Applications

Random number generators (RNGs) are essential to cryptography, simulations and secure digital communication. Unfortunately, most deployed RNGs have no runtime visibility into their quality of randomness. They are “black boxes” – providing numbers with no accompanying service or classification describing entropy strength. In this paper we introduce class based RNG architecture which solves these problems. Numbers are continuously classified in real time into one of four strength grades: Basic, Statistical, Cryptographic, and Military. Classification is determined by minimum p value across a rolling window of NIST SP 800 22 statistical tests. The generator uses two entropy sources (environmental noise + user timing input), a cryptographic extractor (SHA 512), and newly designed Adaptive Nonlinear Scrambler (ANS). The ANS layer breaks linear structure with data dependent feedback paths. Evaluated on 1 million bits of output, this generator scored a minimum p value of 0.82 (Cryptographic grade) on all 15 NIST tests with all tests scoring higher than p = 0.82. Our random number classifier works on real time and has demonstrated 97.1% accuracy identifying correct operating grade and auto triggered reseeding events when entropy grade weakened below Cryptographic strength. Benchmark comparison with three reference generators (Mersenne Twister, Intel RdRand, OpenSSL) revealed our RNG was the only one to pass Linear Complexity and the only one offering runtime grade labelling. One case study has been included for AES 256 key generation.