HAVEGE

"One of Peter's colleagues replaced the random input source employed by HAVEGE with a constant stream of ones. All of the same tests passed."

Um, that seems like a weird test. If we had a machine that output apple pies and claimed to use genuine apples, wouldn't it seem more logical to examine whether real apples were an actual input rather than tasting the output to see if it tasted like apples? Put another way, why wouldn't you examine the data HAVEGE feeds its internal PRNG rather than testing the output of the PRNG? It seems like the test was designed with a certain conclusion in mind.