With that reasoning, one doesn't need an initial randomness file to begin with. Either you think there's plenty of usable entropy soon enough after power on th

With that reasoning, one doesn't need an initial randomness file to begin with. Either you think there's plenty of usable entropy soon enough after power on that values can't be predicted, or you think there isn't and we need to keep a file to initialise from > an attacker which can feed your PRNG non stop with their own entropy Is not needed if you know the initial state from the file (which was present in the image file an attacker got their hands on, for example): knowing a sufficient percentage of outputs (such as the tcp seq/ack number (I forget which one the server generates) when you send a decent rate of syn packets) lets you verify which entropy addition would have lead to that output being generated. That's a handful of bits to brute force at each step unless the entropy additions are processed in prohibitively large batches (say, ≥80 bits at a time) which means you start completely predictable until that entropy is accrued (which may or may not be an issue depending on circumstances)