I also solved part2 manually today, by outputting the correct addition and program output in binary and then reversing the path for all wrong outputs.

Then I just had to compare that to the formulas for a full adder and I found my swaps pretty quickly that way.

I did all of this in Kotlin and on paper, with a handy helper method to construct the paths.

It’s at the very bottom of this file on Github.

I suspect that comparing the bits and then comparing the paths to the full adder formulas can also be done ‘automatically’ decently easily, but I was too lazy to implement that today.