Interesting! So basically validate a Plasma Cash-style Merkle tree inside of the block, so light clients can do a O(log(n)) sized query per block if that’s what they want to do.
This is something I am trying hard to avoid. The point of abstraction is to simplify consensus code and allow consensus nodes to be simple and dumb even if the applications running on top are really complex, but that’s of limited value if the code that block proposers need to run to figure out what transactions to accept to get fees is complex and requires understanding all of these layer-2 schemes anyway.
Relayer markets seem to be the most general-purpose approach unfortunately…