The worry is
shadows, but I'm encouraged by the fact that the MPP voltage shouldn't
change too much in shadows, so the PID loop will handle it. The best
way of doing the sweeps, as you suggested, would be to short the array
(or, in our case, get it as low as possible), and then let the input
caps charge while sampling like the clappers. Since you have to let
the input caps charge when you're doing it the other way around
anyway, this has to be faster.

The efficiency of your models sounds awesome. I wish I'd designed
those! :-). Did you look at Silicon Carbide diodes for the output?
They have no appreciable reverse recovery...

We copied the Drivetek/Biel protection method on our trackers --
there's a few comparators which trip on the maximum input voltage,
minimum logic voltage, and the minimum input current. It would be
pretty hard for the hardware to blow itself up, even if the software
went AWOL. It's saved be a lot of times.

Thinking about it, you must have some kind of start-up or battery-
detection method? Ours just boost to the configured maximum battery
voltage. When the battery is connected, it drops to its actual battery
voltage, the output control loop comes out of saturation, and the
input control loop takes over.