Leela can normally read 25 move ladders. It just takes a while. Positions involving long ladders can be a real drag on Leela’s performance, in my anecdotal opinion 
The AI review is currently broken. You remember the other thread, where you replied directly to my comment about the issue? We’re still waiting for @anoek to get on the case 
Meanwhile, based on a Leela console output screenshot in a PM thread, and my reading of Leela source code, I have begun to suspect that the percentage values that you see on the moves are in fact the direct output of Leela’s neural network. In other words, her “first intuition” on the position. They are often plausible, but essentially worthless.