You got most of it nailed.
Let's say that we're dealing with an irrational actor that cares not for profit or anything else but sheer incentive disruption.
If 51% attacks then they really only at most can control the chain tip at best , if they even try a 1 block reorg they have to re do the work of 10 blocks to even outpace the network working honestly. It's only a 1% advantage.
Check out this episode for more.