Pass/Fail criteria for the Ent suite
I have compiled criteria for passing and failing the Ent tests. They are based on data available at http://www.fourmilab.ch/random/ and at https://sites.google.com/site/astudyofentropy/background-information/the-tests. In the case of three tests, neither source clearly delineated desirable and undesirable values. For these, I assumed that the Yarrow PRNG passed, added a tiny margin of error, and called everything outside of those conditions a fail. This was done for the Arithmetic Mean, Monte Carlo - Pi, and Serial Correlation tests.
Entropy
7.7 and above passes.
Compression
1% or less passes.
Chi-Square “exceed this value” percent.
Less than 1% or greater than 99% fails.
Between 1% and 5% or between 95% and 99% is "Suspect".
Between 5% and 10% or between 90% and 95% is "Almost Suspect".
Between 10% and 90% passes.
Arithmetic Mean
Between 126 and 129 passes.
Monte Carlo - Pi
Error less than 0.07% passes.
Serial Correlation
Correlation smaller than 0.005 (plus or minus) passes.
Pass/Fail criteria for the Diehard suite
Based on the following quote from Dr. George Marsaglia, the developer of the test suite, only tests that return a 0 or a 1 are failures.
"You should not be surprised with occasional p-values near 0 or 1, such as .0012 or .9983. When a bit stream really FAILS BIG, you will get p's of 0 or 1 to six or more places. By all means, do not, as a Statistician might, think that a p < .025 or p> .975 means that the RNG has "failed the test at the .05 level". Such p's happen among the hundreds that DIEHARD produces, even with good RNG's."