[jparkie]

I'm thinking of building a test suite which takes a distribution of distinct elements and run a test for FPP and FNP. Then it asserts whether the the actual FPP and FNP is within an epsilon of the calculated probability.

I want to add statistics for mine to hopefully be able to monitor them at least by some estimate.

[nightcracker]

I would like to add to this that it's incredibly important you use fixed seed, or at least logged seed tests for this.

If you fail to do this you get unreproducable ghost tests that you can never investigate if they happen to fail rarely.