You're right, it does have 14.4 stops of DR, despite a 14-bit ADC. It's also powered by an internal perpetual motion machine, floats in the air when released, and basically defies many other laws of physics and thermodynamics.at the pixel level, the D800 DR at ISO 100 is 13.23EV.
the 14.4 is because of the normalizing in the DXO print mark.
Yes, I know.
If a data analysis method includes a normalization step which forces data to fall outside of the range that's physically possible for the measurement, that data analysis method is flawed, and by extension, any conclusions based on that method are also flawed. If a hospital reported to parents that their newborn infant had a population-normalized length of -4", you'd say WTF, a negative height is impossible, right?
Same thing with a 14.4 DR from a 14-bit ADC. WTF, that's impossible. Change the method, becasue the method is flawed. If the analysis method is flawed, the resulting conclusions (i.e. DxOMark's Scores) are also flawed. Note that I think (and I've repeatedly stated) that their Measurements are valid and useful - it's the Scores, which are based on the flawed normalization step (and have other problems, like undisclosed 'black box' weighting of sub-components) that are meaningless.
IMHO, this is so true. The mass majority doesn't fully understand the testing procedures but only look at the 'numbers' in order to position the worthiness of the product. The real test is using your own combination of body and lens in your own environment with your own light and producing a final print that satisfies your own level of professional expectation.