Thanks for the post @0xcchan.
The formula 8 is incorrect because it does not take into account fees. Uni v2 has a constant 0.3% fee, it should be very easy to that incorporate in the theoretical model. Once you do that, my guess is that Figure 9 will show a much better fit between the PnL and predicted PnL. (Or how do you measure the PnL, do you account for tx fees and payment to the builder?)
I’ll admit that I was lost at Figure 6 and the following discussion. Correct me if I’m wrong - what you plot is the theoretical revenue, computed via price difference in the CEX and DEX. How is it possible that the difference goes down so sharply before the block is published? Only trades can change price in the DEX, and trades cannot happen before the block is published. Also how can one get negative revenue?