"I've done the usual test where you use a factory edge on a sheet of plywood and then mark a line, flip the square, and then mark a second line and look for divergence. Seems like one time a particular square looks correct, then the next time I test it might not look right."
There's one issue with that method. If you register the top of the bar on the square against the test surface and mark then flip it and mark again, you will now have 2 lines to compare. They may or may not be parallel which is the desire condition. If they are NOT parallel, the amount of error is actually only half of the difference. So now you have to correct for that difference by hammering or bending the joint to reduce it.
Here's my point. The top of the bar is your "test" reference, but you have not yet tested the bottom of the bar.
The bar may not have parallel edges for some reason. It's not going to be a huge difference, but the potential is there regardless. The bottom is the part of the square you will use the most often in a cabinet corner to check for square. The top of the bar is used more often in the construction phase.
I would use a marking knife when doing this test because it will be more accurate. Another method is to test all your squares with the bar facing to the left at first. Check to see if all those lines are parallel, testing them against each other. Then you can flip them so the bar is facing to the right and repeat the test. If one of the lines is "off", there is an issue with that square.
Another method is to place the square on a flat surface holding them vertically and butt them together so the blades can be compared one against the other. Check for a sliver of light that can be seen between the edges of the blades. This method checks the bottom of the bar, but only to another square.
WE are splitting hairs here and usually not necessary for woodworking accuracy.