I believe that "refining" is simply a Band-Aid on a faulty profiling process. I agree that this is a major upgrade for 2.0.
-- A lot of speculation below --
A big part of refining may be EQ matching. If you are able to tweak an amp sim until it sounds exactly as close to the real amp as the typical non-refinded profile does, with its inherent faults, and then EQ match with the same tech Kemper uses, the amp sim will get similarly close to the real amp.
But there are often sacrifices with that approach.
Sometimes EQ match just won't bridge the gap too well. It doesn't matter what you do; the differences aren't about EQ alone. They are about other variables... The "meat" of the amp. The "sausages." There's "stuff" EQ doesn't correct for. I see this all the time in my noob tests. No clue what the right words even are, but it doesn't matter.
To get closer, you need to better match the amp sim to these underlying characteristics (the fundamentals, whatever they involve). It could be a different master volume setting, for example. And then, if needed, EQ match is the cherry on top.
So I believe this may be one weakness of profiling 1.0, assuming it works as I imagine. The automation process picks from a bunch of amp sim variables to emulate the amp (gain levels and "sausages"). Then EQ matching tops it off, especially during the refining process.
The possible weakness is this: your underlying amp sims (or big-ass amp sim with a large number of automatically set parameters) may not be close enough to the real amp for these "meat," non-EQ qualities. It may be a "kinda sorta close maybe" approximation that relies heavily on EQ matching, but isn't nuanced like many dedicated amp sims can be (e.g., Fractal, Helix, etc).
And then you're left with a capture that sounds quite close to the amp, and can even be confused with the amp, depending on the blind test... However, finer nuances may not be there, or they get homogenized.
This may be part of what people experience with high-gain Kemper profiles. I believe it is. The underlying amp sim/sims and their parameters tend to feel and sound a specific way, especially in these so-called "nuances" (a loaded term, but anyway). And they simply can't get closer to the real amp.
(Well, they could, with better EQ matching than what profiling 1.0 may do, but that aside.)
That is what CK may have worked on. First, taking a look at the underlying amp sim parameters or multiple sims. Working on things like the low end, and developing or adding parameters that "fix" some of the usual perceived flaws of profiling 1.0. He likely would know what to look for by now.
That could allow Kemper to get closer to real amps right off the bat. Add in automated refining, the processing power of computers (which may be needed for this evolution of profiling), much better EQ matching, and here we are.
30 seconds to produce a result.