To fix this, I'm going to change out some of the parts. To follow along, look at sheet 2 of whatever schematic you like.
The pilot generator has two pairs of transistors, an output pair that switch +12 or -12 to the output, and an input pair that switch the output pair on and off depending on the state of the logic level output from the controller.
The resistors from the emitter to base of each of the output pair will change from 1k to 47k. The resistors from the base of the output pair to the collector of the input pair will remain at 10k. The resistors from the base of the input pair to the pull-up will change to 10k, and the pull-up will change to 1k.
This will result in a decrease in the collector current of the input pair coupled with an increase in the base current. The hoped-for result will be a more solid, assured entry into saturation for the input pair. My thinking is that the problems some folks are having are centered around incomplete turn-on of the input pair, which results in nothing on the output pair for either + or - 12, which winds up causing diode check errors or invalid states, or cars being upset with the pilot or what-not.
Additionally, I have a new design which will replace the four transistors and their bias resistors with two pre-biased complementary pair transistor modules. The whole thing will shrink down to 2 TSSOP-6 packages and a single pull-up resistor.
Also, another revision I've decided to make is to go back to the 20k (instead of 16k) resistor value for the low side of the voltage divider for the GFI comparator.