You keep repeating it, but it's still wrong. It's wrong because the low end of the dynamic range is determined by SNR, and SNR changes when you downsample.
This is the belief of a math geek who spends too much time with graphs and not enough time with real photographs. Down sampling throws away noise and detail. Dynamic range is the range of useful photographic detail. It is not the simple difference between two numbers on a graph.
Of course it throws away noise AND detail, the fact that there is a noise-detail tradeoff is the point !
The point is that you normalize all cameras to the same resolution so that all cameras are at the same level of detail
. The test is, how much dynamic range does each camera have for some given resolution
I understand that there's a whole other can of worms regarding at what point "useful" dynamic range starts -- is it SNR = 0db, 10db, or maybe even 20db ? But I don't think that's what you are referring to.
It doesn't make any sense to compare data points which represent different tradeoffs/choices on the noise-detail continuum. Now, to your example --
Try this: shoot 4x5 Velvia. Drum scan it. Down sample to 10 MP. Did you all of a sudden gain 3-4 stops of shadow detail? No? Hmmm...
Here's a different test for you -- sample it at 10MP, create another image at 40MP. Then compute the "blackpoint" (SNR = 0db) of the two images.
Now I take it that we agree that the images both have the same "true" dynamic range, but I put it to you that the measured blackpoint on the 40MP will be at a higher luminance level, and therefore the measured per-pixel dynamic range
will be lower on the 40MP image.
Without normalization, your measured dynamic range has more to do with how the data was sampled and less with this useful photographic detail
that you're so fond of.
Shoot a Stouffer transmission step wedge and look at it with your own two eyes. That tells you what you can expect in the real world.
For what it's worth, I am in agreement with you here -- if a test site were to include such shots, and allow the user to judge, it would be quite valuable. Of course it wouldn't resolve the debate about whether to "normalize" to a common resolution (or for that matter, what to do about different ISOs).