It depends on how you write it. In my version, every tapleaf commits to a single logic gate and they are all nearly identical in size, except for OP_NOT (also known as an INVERT gate) and OP_LID (also known as a BUFFER gate), which are considerably smaller than the other logic gates because they only have one input and one output, which means they only have two input hashlocks rather than four.

However, Robin's model relies on a very different architecture than mine. One reason why is that he tries to do as much "work" in each tapleaf as he can so that bisection happens in fewer transactions. Consequently, his version is more efficient than mine.

The 8 bit cpu I wrote for bitvm can square and multiply any 8 bit number and keep track of overflows. Robin's version has a 32 bit cpu that can do the same on 32 bit numbers. I don't think we are far from being able to do key tweaking.

I was too harsh on Vicky, she is a wonderful and integral part of making bitvm work. Still, I'd like to eliminate the role of the verifier someday because bitvm would have better trust assumptions if Paul was "unable" to be dishonest rather than just "it's expensive to be dishonest if Vicky catches you."