[Klaus]

I get that if you are interested in LENS performance you can compare numbers within a system, or to some degree charts.

 

My question was whether if you are interested in SYSTEM performance (i.e. lens+sensor) you can compare raw numbers. Thus if the we see the same numbers for a lens on a 16MP system an a different lens on  36 MP system we can infer that we will see similar resolution in a print of the same size (or if we normalise the two images and view on screen). 

 

This is supposed to be how perceptual megapixels work with DXO; if one lens+sensor gives you 13PMP at a certain location and a different lens+sensor gives you 13PMP they will look the same printed to the same size (at that location). Of course if the first sensor is 16MP and the second 36 you might want to say the second lens is 'bad' - it isn't giving you the potential of the 36MP sensor. But it's still giving you there same performance in one sense as the much better lens on the smaller MP sensor.

 

In some way these are the numbers that are of most interest: to me, for example, in figuring out if the a lens and sensor from one system can give the same appearance of sharpness in a certain print size and a different lens mixed with a different sensor.

 

 

The tests are not aligned to this purpose. You can do so but in this specific scope the error margin is probably higher than over at DxO. e.g. we are using the RAW converter that yields the best results for the specific system whereas DxO probably just sticks to their own algorithm. That being said, the problem remains that e.g. demosaicing algorithms tend evolve over time. e.g. for Fuji X-Trans Adobe's first shot was rather poor but improved significantly with later versions. No idea how DxO handles these life-cycle dependencies.