Indeed, trying to aggregate chunks is definitely in the direction of “search within DoS-compatible solutions”, which I believe is the wrong way to go. If we’d have spare cycles, and somehow considered doing something more computationally expensive, it should be searching aggregates of chunks (or subsets of chunks) for better incentive-compatibility - not DoS-compatibility.
Continuing my thought from yesterday, the example with a high-feerate small parent chunk that evicts more than it pays for and low-feerate child that would pay for it. If this is actually a goal, instead of raising the feerate of the child chunk, one can also just attach a small high-fee additional transaction to the parent chunk. So perhaps that makes it not that much of a concern (even if a use case exists, which I’m not convinced about).
So overall, let’s stick with chunk per chunk.
What if it’s a failure due to recent softfork or so?