it's much easier to talk and reason about these systems if we focus on the end-properties we want

we don't want decentralization for its *own* sake, we use decentralization to achieve a specific property: hardness.

i.e. a durable, credible guarantee that "this system will continue to behave predictably in the future" and will not be captured or controlled by anyone, resisting censorship and other manipulation. This is the foundation of what makes blockchains valuable.

decentralization of various kinds ("anyone can run a node and there are thousands of them", "updates to the protocol are made by a large community in the open", etc) is a core part of how that is achieved on L1, but it is not the only tool we have. Cryptography also gives us tools to set hard constraints to prevent manipulation and enforce guarantees.

for instance, in a ZK rollup the requirement that a centralized sequencer submit a cryptographic proof of each state transition means that, even though it is a "centralized" party, there are hard limits on how it can use that position to harm users.

the hard guarantees offered to users is the point, and sometimes "naive decentralization of every layer of the stack" is not the best way to get there.