So the fun part with 4 channels was it wouldn't go full off. Actually there's a note about that problem already for the 3rd channel (runs off the same PWM clock), but I didn't understand it until now. Ok, so only phase mode goes full off and timer 1 doesn't do phase mode. The 3rd channel historically drives the FET though and isn't used for moonlight on tripples, so probably just never noticed. Also it's likely significantly worse using interrupts to bit bang the pin. The simulator shows about a 6 pwm-tick delay between the two interrupts handlers running, which could just be from the interrupt handling itself. The first interrupt already starts 3 ticks after the match, so maybe that adds up. I doubt the wave form generator is that bad. I think it's supposed to be a 1 tick spike. Anyway, I'm using the interrupts for my idle sleep clock but I sent a control bit to the on intterrupt to just shut it off on level zero and it behaves. Whoo. I haven't done the same for channel 3, but probably should for potential RGB use.
TK had a note awhile back about not knowing how to slow down timer1 (the channel3/4 timer) below 31 khz. In principle that helps some with this issue and generally with low modes. I'm not sure why she said that though. Actually before I read that, I thought I'd slowed mine down by half. It has a prescaler (TCCR1 CSx), which I thought I adjusted, but now I'm wondering if it doesn't work, or if I was confused. All my blink timings are based on that clock when the 4th channel is engaged so I'll just play with it and see how the blinks go and find out.
E-switch works, but there are many configuration possibilities with slightly different code paths, and the one I tested is maybe not even very normal (ok, tested that one too, it works!), so a little more to test still. Well, there probably is more testing to do, but I'll probably go and and put it out in the next day or so anyway.