An optimistic rollup is, in essence, a contract: a piece of code on L1 that pretends, for a while, that whatever the sequencer says is true. Transactions are executed off-chain by the sequencer, their effects compressed and posted back as state roots, and a window opens during which any honest party may submit a fraud proof. If no one objects, the state becomes canonical.
The economic logic is elegant. Most blocks are honest, so most blocks need no proof — only their data is published.
The cost of running an L2 thus collapses to the cost of data availability: posting enough information that anyone,
given the will, could reconstruct the chain and challenge it. The challenge window, typically seven days,
is the price the user pays for low fees in the common case.
The trouble is liveness. If the sequencer goes silent, withdrawals stall. If the data is withheld, fraud proofs become impossible. Every detail of the design — from the choice of DA layer to the inclusion-list mechanism — is a hedge against this single failure.
Three pillars
- Execution. The sequencer runs the EVM (or a variant) and produces blocks at low latency.
- Settlement. Periodically, state roots are anchored on L1 along with input data.
- Dispute. Any verifier may run the same execution and submit a fraud proof if the posted root is wrong.
Most blocks are honest, so most blocks need no proof — only their data is published.
What survives this design is a chain whose security is, in the limit, that of the host. What is gained is throughput. What is lost, or rather deferred, is finality — an L2 transaction is final to its sequencer immediately, final to L1 only after the dispute window closes. The art of L2 design is in collapsing this gap without weakening it.