Originally posted by rrstuff DXOMark has a nice methodical approach, but they have three small issues. First is lack of inclusion of metamerism index in the score* and the second is lack of measuring the resolution. The first is important because of overall color quality and you can sacrifice it to get a bit better iso performance. The second could account for noise filtering in RAWs, which, when applied, would yield a higher ISO score, but not better image quality. Third, is the grain size. Classic example of how it mattered would be original four-thirds cameras. Their sensors had low noise, but larger grain and the pictures looked terrible at high iso, despite having relatively low noise, mathematically speaking. Old nikons had quite a bit of noise, but they had small grain and people liked the way images looked.
I agree DxO has minor issues in their methodology. But none of those you mentioned.
1. Metamerism: It is checked to remain above some threshold DxO says above which differences remain invisible. I agree it should influence the score though.
However, you cannot sacrifice metamerism to get a better score. True, broader (more white) filters reduce luminance noise. But they increase color noise and worsen the color depth score. It's a rather clever balance actually.
2. Resolution: It is printed bold, no need to check #pixels of a sensor.
DxO checks for noise being white noise (i.e., with no raw processing taking place and that there can't be a raw grain size, always 1 px). All noise blocks you see in images are from raw demosaicing and/or JPG processing. DxO tests before demosaicing.
One of the true minor issues for instance is that DxO does not test against fixed pattern noise in scaled down images, i.e., they don't verify their scaling law which only applies to random white noise.