Thanks for your response!
In the model we are using, we assume that blocks the validity of a block depends on the block being proved on time – if you build, it you prove it as described in Prover Killers Killer: You Build it, You Prove it. IMO prover killers aren’t an important consideration here, because if the block isn’t proved, the person who built it suffers the economic consequences of not collecting the rewards from the block becoming canonical.
This paragraph does point out an interesting feature, which is that our model assumes all of the provers have a homogeneous cost of producing the proof (while this might seem simplistic, it is still remarkably difficult to say anything about this model under an adversary, which is what we are trying to do now). Further, we don’t explicitly model the repeated version of the game where provers have to make their proving decisions in the context of future expectations on proving rewards. Clearly there is more meat on that research bone, and in the full version of the work, we hope to justify the modeling decisions we make by striking a balance between tractability of the model and the most believable assumptions.