That is a gross misrepresentation of what I am trying to say. If you actually read the material that I linked to, you would notice that it affirms my claim that even well recognised sources of randomness must be tested to check that they are suitable for use as a RNG in cryptography. If they are not suitable as implemented, then it may be possible to account for bias, but failure to do so means that the resulting RNG's output cannot be used for a one time pad.Originally Posted by Mario F.
But there's more: I am also concerned with a practical application of this RNG, given that it depends on source text that might have a tendency to be repeated. It would be more difficult to test for this though, since this depends on the implementation and patterns of human behaviour should humans be involved in source text selection. But if humans are not involved, then how can this be correctly automated?