What is known about the randomization quality?

[daniel-vainsencher]

"No guarantees" is pretty discouraging... is this using a published algorithm? is there a high-level overview of what it does?

[TimNN]

I mainly wrote this for a toy project which has by now been scraped.

When testing the code, the generated sequences "looked" random enough to me, however I never tried running it through some kind of randomness evaluator.

Internally it essentially uses a block cipher with a randomised key, however I could not find a published algorithm which supported sufficiently small / variable block sizes which is why I had to roll my own.

The actual algorithm used is defined here:

https://github.com/TimNN/shuffled-iter/blob/master/src/lib.rs#L106-L108

and basically looks like

next_value = ((((previous_value * f1) ^ x1) * f2) ^ x2) & mask

Where x1 and x2 are arbitrary numbers, & mask is essentially modulo 2^n and f1 and f2 are factors for which a multiplicative inverse modulo 2^n exists.