So Urbit's Layer 2 solution is a Rollup, but not an Optimistic (fraud proof) or ZK (validity proof) Rollup, but their own homespun form of Rollup that they call a Naive Rollup.
It's described here:
https://18108973658826589741.googlegroups.com/attach/31d6ad2...
It appears to use Ethereum L1 solely for data availability and sorting, while the computation that ensures validity is performed inside Urbit instances. I haven't thought through the security implications of this yet.
I believe the implication is that you need to run a full Naive Rollup node, that validates all Urbit transactions, in order to ensure the validity of any given transaction.