Each archetype is the same five layers — consensus, execution, data availability, settlement, proof — assigned to different actors. Where the boundary falls determines who you trust.
One chain handles consensus, execution, data availability, and settlement together.
Performance comes from putting everything in one place. Trust is reduced to: enough validators are honest. Re-execution is the only proof — there is no proof artifact.
Layers are sold separately. Each rented from a different provider.
Specialization. Each layer can be replaced. The cost is more interfaces and more attack surface — bridges between the layers themselves become a trust point.
Execute optimistically. Trust 1-of-N watchers to spot a fraud and submit proof.
Cheaper than ZK rollups. The 7-day challenge window is the price. Liveness depends on at least one honest watcher being online with a working bridge to the L1.
Every batch is accompanied by a validity proof verified on the settlement chain.
No challenge window. Withdrawal is final the moment a proof is verified on L1. Cost: heavy proving, often a trusted setup, sometimes a centralized sequencer for the moment of submission.
Sovereign L1 with native ZK. Cross-chain anchored to EVM/TRON via on-L1 verifiers.
Not a rollup — Psy is its own L1. But its bridges are settled by ZK proofs on EVM and TRON, so withdrawal trust reduces to circuit soundness, not validator honesty. M-of-N relayers exist for liveness only — any user can self-generate a withdrawal proof if relayers go silent.
The layer's safety property is enforced by math. No human can override it.
Distributed trust. Many actors must collude to violate it.
A small set of actors can violate the layer's safety. This is where attacks land.