I promise on my family's graves three generations back that the supplied files are exactly as I have suggested (within the limits of integer math - for instance if a sample of noise is at 5/65536 of full scale and you half it, it'll be rounded up to 3/65536 since digital samples are never expressed in floating point math) but obvious caveats aside, this is a more than fair assessment of all these numbers and discrepancies everyone throws around but hardly anyone understands.

Bren R.