It occurs to me that there may be a way to demonstrate that a simple mod that costs essentially nothing will do more for this speaker than all of the improvements in the version that is soon to be shipped.
Ctrl gave us convincing reason to believe that the peak in the off-axis response above the crossover point is due partly to diffraction, exacerbating the directivity mismatch. When he doubled the baffle width of a similar speaker he was modeling in simulation software, thus forcing the diffraction ripple to shift to half-lower frequency, the response peak above the crossover point was reduced in magnitude by about half.
It follows that any modification that is effective at suppressing the diffraction ripple in the off-axis response should have a similar effect, although not likely as strong. Studies of the effect of wool felt have shown it to be effective at reducing diffraction. Genuine wool felt, with thickness 1/2" or more, placed strategically to either side of the tweeter, with height matching the diameter of the tweeter flange and extending from the lateral edge of the flange to the start of the baffle edge roundoff, will likely suppress diffraction to an extent sufficient to be discernible in the measurements.
A comparison of the version already tested to the version soon to arrive will likely show that rounding the baffle edge hasn't any appreciable effect at smoothing out the off-axis response in the vicinity of the crossover.
A comparison using either version (preferably the one soon to arrive) with and without the use of wool felt as described above will possibly show that the felt is more effective than the edge rounding at suppressing the diffraction, and will possibly show that this simple inexpensive mod is more effective at improving this speaker than all the costly mods that have been applied to it.