This is purely a technical question.
Profiling works by the Kemper feeding a signal into an amp and listening to the responses to determine how to set the parameters of an internal model. I always presumed that the Kemper sweeps through the entire range of frequencies that guitars can produce and multiple volume levels to get a complete picture of how the amp will respond to various inputs.
So why would subsequently feeding it a random guitar signal -- which should be less comprehensive coverage of amplitudes/frequencies than the original profiling signal -- make the profile better?
Is it that the original profiling signal is less comprehensive than I imagined, and the guitar is filling in the blanks? Because that would indicate that a refined profile is only ever going to be truly accurate with that specific guitar. Is that the case?
Why can't the original profiling signal be more comprehensive, so that it covers every possible guitar signal?
I'm willing to hear people's speculation, but I'd be especially interested to hear if Kemper himself has every commented on this particular question.