Layer-two networks move transaction execution off the base chain; what remains on-chain depends on the design.
The useful choice is therefore not “on-chain or off-chain,” but which work moves. An L2 runs a separate execution environment where transactions update balances and smart-contract state, then sends a compressed record, state commitment, or proof back to its base layer.
What the base layer still does
For a rollup, the base chain remains the settlement and verification layer. An optimistic rollup posts transaction data and assumes its state update is correct until someone challenges it with a fraud proof. A zero-knowledge rollup posts a validity proof showing that its batch was executed correctly. In both cases, data availability matters: independent participants need enough information to reconstruct the L2 state or verify withdrawals.
That is why “off-chain” does not mean “unaccountable.” The L2 sequencer may order transactions and execute them quickly, but the base-layer contracts decide how deposits, state roots, proofs, and disputes are accepted.
When building on an L2 makes sense
Build on one when its execution environment and settlement guarantees fit the application, not merely because its fees are lower.
- Check whether its VM supports your contracts and tooling.
- Identify who operates the sequencer and how censorship is handled.
- Confirm where transaction data is stored and who can retrieve it.
- Measure finality, withdrawal conditions, and cross-chain messaging support.
Messaging is separate from execution. A network such as Manta Atlantic may use Cross-Consensus Messaging, or XCM, to communicate across systems connected through Polkadot Network; that does not make every message an L2 transaction. When an asset itself must move between networks, the deposit-and-release mechanics are the separate concern addressed by the Manta Bridge guide.