I think DeepCrypto would agree that any meaningful solution will/must involve calculating difficulty (or difficulties) by some measure of both target time and block acceptance frequency, somehow, and must scale both.
Calculation difficulty shall prevent map preselection cheat. I am not sure that any other difficulty measure would be needed if map will be big enough to force competitors to actually play the game.
And yes. Time warp bug shall be closed somehow anyway. Time warp attack is not very relevant to anti bot measures.
I meant as a solution to the time warp bug. Basically, I think we can say that block acceptance frequency needs to be constrained/controlled directly by the network via the work function's complexity (not indirectly by how people(/bots) are assumed to play) and until it is it still potentially leaves the attack vector open. With any solution that only constrains based on target time any new improvement in machine mining that better calculates bad (high TT) solutions could rewrite back to the last checkpoint with a longer, lower difficulty chain. In other words, if we are only ever constrained by target time then anyone who could just find a way to run the naive bots much much faster could dominate production and/or "51% attack" by difficulty-time-warping the chain, and "should" just not bother with any AI. This would be an undesirable outcome for everyone.