I made a mistake today and fried another TIP 35 or 36 transistor, they don't seem very tough to me, maybe I should have parallelled 2 or 3 of them, oh well, I think this design should work all right, it's similar, but improved from what I had before.
One strange thing I ran into with spice was in the + half of the power supply design, I used a dual opamp and only needed one, so I used the other one as a buffer and if I include it in the simulation, spice finds single matrixes in other areas of the design. Another strange thing is that if I try to sim with a LM317 and a LM33, it won't sim the LM337 half of the design. It will if I delete the LM317 though.
I have some LM317s that are rated for 1.5 amps and some for .1 amps, if you look at my drawing, that's why the LM317s are labelled. This new design appears to be very stabile, as you put more load on it the voltage stays constant.
I did find where Audioguru was correct, when there was no load and the dial was set for 0, it would put 4 to 8 volts out and drop back to .7 when loaded. To keep it from floating up, a 250 resistor from out to ground seems to work, but it needs to be a 2 watt for when the dial is set to 20v. Another nice thing about setting the compare voltage, 10V, with a lm317, .1 amp, is you can tweek the output so that when I set the dial at 5, I get 10.0 volts and at 7.5 setting, I get 15 volts.
I hope to finish it up tomorrow, again, and I'll take a few photos of it.