I think I'm missing the point of what you're saying here so I was hoping that if I explained why I don't understand, perhaps you could clarify.

VNM-utility is unique up to a positive linear transformation. When a utility function is unique up to a positive linear transformation, it is an interval (/cardinal scale). So VNM-utility is an interval scale.

This is the standard story about VNM-utility (which is to say, I'm not claiming this because it seems right to me but rather because this is the accepted mainstream view of VNM-utility). Given that this is a simple mathematical property, I presume the mainstream view will be correct.

So if your comment is correct in terms of the presentation in the FAQ then either we've failed to correctly define VNM-utility or we've failed to correctly define interval scales in accordance with the mainstream way of doing so (or, I've missed something). Are you able to pinpoint which of these you think has happened?

One final comment. I don't see why ratios (a-b)/|c-d| aren't meaningful. For these to be meaningful, it seems to me that it would need to be that [(La+k)-(Lb+k)]/[(Lc+k)-(Ld+k)] = (a-b)/(c-d) for all L and K (as VNM-utilities are unique up to a positive linear transformation) and it seems clear enough that this will be the case:

[(La+k)-(Lb+k)]/[(Lc+k)-(Ld+k)] = [L(a-b)]/[L(c-d)] = (a-b)/(c-d)

Again, could you clarify what I'm missing (I'm weaker on axiomatizations of decision theory than I am on other aspects of decision theory and you're a mathematician so I'm perfectly willing to accept that I'm missing something but it'd be great if you could explain what it is)?