To apply the physical lens profile model to an image, ACR needs to know the physical pixel size of the image. In most cases, ACR does this by looking for available EXIF metadata (prefers FocalPlane*Resolution tag, but also uses combo of FocalLength and FocalLengthIn35mm if necessary). For some cameras, ACR has an internal lookup table to deal with known bad metadata cases (e.g., I've seen some compact models with bogus metadata values, such as claiming they have 12-micron pixels, which is ... not right). But for the most part, the calculation of the pixel size is done from metadata.
The SFF data in the profile refers to the source data (i.e., the size of the image circle represented by the lens), rather than the destination image to which the profile is applied. This is because for interchangeable lens systems (e.g., SLR), the lens may be for 35 mm format whereas the destination image may be smaller format (e.g., APS-C).
In your case, where the metadata describing the pixel size of the destination image is not available, ACR/LR should fall back to using the SFF as described in the profile. That is, ACR/LR assumes that the physical image circle size described in the lens profile exactly matches that of the target image. But it sounds like that's not happening in your case, and I don't know why. If you can provide me with a sample image and corresponding profile, I can investigate (you have my email addr, I believe).
Hi, Eric, thanks for the quick answer
Yes, in the meantime I studied this stuff a bit further and concluded that reading the crop factor from raw metadata instead from the profile is indeed the right way, for the reason you specified (interchangeable lens systems)
So in my case I obviously have to correct those metadata (and rescale the profile parameters accordingly)
In case you want to investigate why ACR don't read SFF from the profile if FocalPlane*Resolution tags are missing, I'm sending you test dngs. One with those tags and one with those tags removed, and the lens profile (which was, however, tuned to wrong crop factor of about 3.7 because of wrong metadata in the dng). However, I think it's not necessary to bother with this, unless you think it is important