If you want to make an absolute flat then you need to measure the distance between the astigmatic foci of the spherical mirror as they are bounced at grazing incidence of the flat under test. You'll need a pinhole source for that. If you are making a flat for autocollimation testing of telescopes, you can have many waves of power ( curvature ) but you must have a truly spherical surface.
All you need is a good sphere of at least half the diameter of your flat. At this ratio an angle of incidence of sixty degrees is required to see the whole diameter of the flat. The sensitivity is the same as if you were testing the sphere on it's own, hence the slower the sphere the more sensitivity you'll have in seeing errors. Mine was F10 at at the knife edge from memory. If you use a larger sphere you'll have a slight increase in sensitivity due to the smaller angle of incidence. Remember that the angle of incidence desensitises the double passing of the light path on the flat.
If you are using a Ronchi grating to test for zones, put the horizontal accross the mirror, otherwise youll get confused as the shape will appear torroidal in anything but the central radial chord. Any zones you see are treated in the normal way. I like the knife edge better as you can `focus' the knife edge to present the surface that is easiest to work on, just as you can with any null test. You get a slight distortion of the position of the zones due to a foreshortening effect; this can be alleviated with a pin stick on the spherical mirror.
Making a flat this way, as observed by Ritchey in his original article, is know more difficult that making a sphere. If you want to make an absolute flat, then this method should still work fine. BTW you'll need a bright light source because of all the reflections, I used a laser which was fine with all uncoated surfaces.
Hope this is of use.
Mark