Very close IMHO, and I'd ping Steve Upton but I'd reappraise this subtly:
The Evaluate Profile Proofing is measuring the errors round tripping through the profile path (if you will).
Since I made some nice conversion routines in Matlab I've been looking at some of the rounding/truncating errors introduced by the 3D LUTs. The biggest errors are on the BtoA side because the PCS is in Lab and the transition to printer RGB space has significant, unavoidable errors at the gamut boundary. The LUT cubes have a resolution of 256/N where N is one the number of cubes along one dimension. For N=36, which is the "high" option in I1Profiler, there is a jump of 7 units in L, a or b. This creates pretty much an intrinsic error of around 3 to 4 dE. OTOH, the AtoB LUTs don't have that issue because the RGB channels are 0-255 and the end points don't go negative or over 255. (actually it's done in 16 bits but easier to think of it in 8 ). So there is some rounding error to the 8 bit drivers but that's quite small in dE terms.
When I look at round trip errors but limit colors to within the gamut where the PCS isn't clipping the results are pretty consistently excellent though actual printed colors vary quite a bit more as a result of instrumentation, print head, limited profile color set, etc.
What I wound up doing is using Patchtool to create a set of patches with small dEs (< .5) which tosses the anomalous BoA cube size created boundary errors then printing and measuring those to come up with statistics on the profile accuracy.