Ask HN: What is the best test for statistical randomness?

Current choices seem to be NIST SP-800-22, DieHarder and Test U01, all of which certainly overlap.

Why is there no battery of tests that includes standards developed by different countries and their obvious biases?

5 comments

[ 4.2 ms ] story [ 24.3 ms ] thread

Lossless compression algorithms. True randomness should yield near 0% compression ratios.

Your partner for a credit fast, sure and adapted to your needs. Your partner on a daily basis.We are proud to help people from all walks of life, no matter what their financial situation might be. https://www.carroncredits.com/en/index.html

aappleby 8y ago ↗

Broad statistical tests are good enough for most all non-crypto RNGs - once you pass DieHarder you're probably good enough for practially all common RNG uses.

Breaking crypto RNGs is a subtle enough art that having a standardized set of tests wouldn't actually prove anything useful - it would be like saying "We tried every key in our collection on this new padlock and none of them worked, therefore the padlock is secure."

-Austin, author of SMHasher

janwas 8y ago ↗

PractRand is a useful test suite.

potus_putin 8y ago ↗

http://random.mat.sbg.ac.at/tests/ "On this page, you will find information on tests for uniform random number generators (RNGs)"