With this I definitely agree.
Actually not even with this I agree entirely, I may add a section for this, but it’s a little beyond the scope of this short note because it delves in different forkchoice design decisions. In the forkchoice implementation that I have in mind, honest validators will consider the vote of the committee of N towards the node of N only if they agree with the PTC view of the block. The attacker may be able to do a vanilla ex-anti reorg with many consecutive blocks in a row, but only if he’s voting on empty payloads which makes the reorg not that interesting of an attack. I do agree that numbers like 20% for PB make these long ex-anti reorgs possible situations more frequent, but I believe we have room here to leverage the honestly of the PTC.
I didn’t get these numbers. If the proposer holds \beta that withholds for slot N targets a x/1-x-\beta split and the builder colludes revealing then the next proposer will base on N-1 if
1- x - \beta > x + RB \Leftrightarrow RB < 1 - 2x - \beta
(We also have x < \frac{1 - \beta}{2})
The block is reorged if :
RB + \beta + x > PB + 1 - x -\beta \Leftrightarrow RB > PB + 1 - 2x - 2\beta
Combining we get 1 - 2x -\beta > PB + 1- 2x - 2\beta
From there PB < \beta which is independent of x.
Perhaps I’m not understanding your attack though.