It is normal for a MPA to skip orders if it detects that the difference between runs is nowhere close enough for converging with the next order. I suspect the idea being to save you time in running the analysis by skipping orders that won't help much. I wouldn't be surprised if that is the case for the runs that didn't converge and never used the maximum order setting. If nearly every element and edge hasn't converged, there's almost no way on earth that the next run will converge within 1%. I don't know what the criteria is for Simulate to make these kinds of decisions but that is the reason behind only selecting the maximum and minimum orders. Let their algorythm do it's job to optimize the time spent on the analysis.
You might want to do a plot of the element orders in the model so you can see if there is something in the geometry that the automesher needs help with by manually increasing the number of elements in a particular area using AutoGEM controls, adjusting the geometry, tolerance, etc.
Depending on the geometry and discretization of your part, 1st order and perhaps 2nd order runs can be useless and even misleading...particularly for the p-method which uses less elements....unless, of course, that is your aim. The default minimum is 3rd order and I'm usually inclined to leave it there so that the shape functions can accurately depect the geometry.
Finally, are you only converging on the max vm stress measure? That could be misleading and dangerous, espcially if you run into a nasty singularity in another analysis...doesn't appear to be the case this time. Is there a particular reason you didn't want the default options of local displacement, strain energy, and RMS stress? Ultimately, you might want to include more measures for convergence. In the report, is looks like max plane stress values for convergence are high even if the values are low. Displacement in Y also didn't converge to your target of 1% (which might be too strict). Just my $0.02, hope that helps.