That's a good explanation on the previous page JR, thanks for posting that for those who need it.
The only thing I'd add to that even if we know the RMS (or averaged) read noise numbers used for their base DR calculation, we would still have no clue as to HOW UGLY lifted shadows will look. That's what pushed-4-stops-lens-cap-on shots are for!
If DxOmark would publish the peak-to-peak read noise, and they'd still likely have to do a compromise of that, or add a read-noise histogram to be more thorough, then we would have a more clear indicator of whether that noise is random in nature, and therefore subjectively less bothersome, or patterned, and therefore more difficult to mask.
E.g. In the case of the 7Ds I looked at, alternating vertical bands of (8 per, i think) pixels had different levels of read noise and this produces obvious stripe artefacts when lifting shadow areas. Since the ratio of read noise to signal there is already very low the read noise difference is made obvious.
In some other cameras the read noise is more uniformly distributed and is less obvious when lifting shadows.
Imaging Resource provides a different way of presenting the DR measurement, by providing DR figures at different signal to noise ratios, still likely based on RMS read noise, but it provides more of a spread of figures which can be helpful and they also publish more noise data as well, including individual RGB curves.
Here they are for the 7D, and the tiny-sensored Pentax Q.www.imaging-resource.com/PRODS/E7D/Z00100_ALO_OFF_acrauto_Step_2.gifwww.imaging-resource.com/PRODS/PENTAXQ/Z0125_ACRauto_Step_2.png
Even I'll say I'd take the 7D's file over the Q's file for lower overall (BER) noise on a real image - at least until I wanted to lift shadows, where the tiny Q's lack of FPN can then show the 7D how that's done.http://a2bart.com/tech/allcamdknz.htm
Then throw in unit-to-unit variability and improvements that may be made during the mfg run of a given body and there are more things to argue about. For instance, my late model 40D, a model initially reputed to have some noise issues, provides noticeably less FPN in pushed files than my early production 7D or early 5D2 did, yet some claim noisy 40Ds and clean 7Ds or 5D2s.
So, back to DxOmark - or any site that does not define how they make those measurements or what baselines they use, even the uninterpretted the data provided is less than ideal or conclusive. (But it's still more informative than "real world pictures" for such matters)
The best rough data from DxOmark on this is the FULL SNR curves because you can see how each sensor performs at each ISO and at a lot of different levels from dark to white levels.