You keep seeing the words — rollup, validium, optimium, "modular DA," "settles to Ethereum." They sound like marketing, but they describe a real, consequential choice every L2 makes about where your transaction data lives and how the chain proves it did the right thing. Those two decisions determine how cheap the chain is, how strong its security is, and — the part that actually matters when things go wrong — whether you could recover your funds if the operator vanished. Here's the taxonomy without the jargon, and why a developer should care.
Two independent questions
Every L2 answers two separate questions. People conflate them, but they're orthogonal:
-
How does it prove its state is correct?
- Validity proofs (ZK): every state update ships with a cryptographic proof that it's correct. Ethereum verifies the proof; if it's valid, the state is final. No waiting.
- Fraud proofs (optimistic): state updates are assumed correct, and there's a challenge window during which anyone can submit a proof that they're wrong. If no one challenges, it's accepted.
-
Where does the data availability (DA) live?
- On Ethereum: the transaction data is posted to Ethereum (as calldata or, since Dencun, blobs). Anyone can reconstruct the full L2 state from Ethereum alone.
- Off Ethereum: the data is posted somewhere else — a separate DA layer like Celestia, EigenDA, or a committee — and Ethereum only sees a commitment to it.
The DA question is the one that names the category. DA is the thing that lets anyone rebuild the chain and exit their funds without the operator's cooperation. That's why "where the data lives" is the load-bearing decision.
The four combinations
Cross those two axes and you get the taxonomy:
| DA on Ethereum | DA off Ethereum | |
|---|---|---|
| Validity (ZK) proofs | ZK Rollup | Validium |
| Fraud (optimistic) proofs | Optimistic Rollup | Optimium |
- ZK Rollup — proofs and data on Ethereum. The strongest security profile: correctness is proven, and data is recoverable from L1. (e.g. zkSync Era, Scroll, Linea, Starknet.)
- Optimistic Rollup — data on Ethereum, correctness enforced by a fraud-proof challenge window. Cheaper historically, at the cost of a ~7-day withdrawal delay. (e.g. Arbitrum, Base, Optimism, Blast.)
- Validium — validity proofs, but data kept off Ethereum. Very cheap, and still cryptographically proven correct — but data availability now depends on the external layer, not Ethereum. (e.g. Manta Pacific posts DA to Celestia.)
- Optimium — fraud proofs and off-chain data. The cheapest, and the weakest guarantees: you're trusting both an honest challenger and an external DA layer.
The word "rollup" specifically means data is on Ethereum. The moment a chain moves DA off Ethereum, it's technically a validium or optimium — not a rollup — no matter how it markets itself.
Why the DA choice is the one that bites
Here's the scenario that makes it concrete. Suppose the L2's sequencer/operator goes offline permanently or turns malicious. Can you still get your money out?
- On a rollup (data on Ethereum): yes. Because the full data is on L1, anyone can reconstruct the latest L2 state and you can force a withdrawal directly through the L1 contracts. Your funds don't depend on the operator being alive.
- On a validium/optimium (data off Ethereum): only if the external DA layer still has the data and will serve it. If that data becomes unavailable, the L1 contract knows a commitment to your balance but not the balance itself — and you may be unable to prove what you're owed. That's the data withholding risk, and it's the fundamental trade-off you accept for lower fees.
This isn't hypothetical hand-wringing — it's the exact reason the community draws a hard line between "rollup" (Ethereum-grade DA) and "validium" (borrowed DA). Modern DA layers like Celestia are designed specifically to make off-chain DA trustworthy, so a validium is a reasonable, deliberate engineering choice — but it is a different security assumption than a rollup, and you should know which one you're building on.
What it means for you as a developer
From the RPC, all four types look nearly identical — standard eth_*, viem/ethers work — so the differences are in how you reason, not what you call:
- Finality timing. ZK types can reach hard finality fast (proof verified); optimistic types make you wait out the challenge window for L1 settlement. If you move value L2→L1, this delay is real — see soft vs. hard finality.
- Reorg behavior at the head. All of them still have a sequencer ordering transactions, and the L2 head can reorg before settlement — so key your indexer on block hash regardless (handling chain reorgs).
- Long-range trust. For anything custodial or high-value, know your chain's DA model. A rollup gives you Ethereum-grade data recoverability; a validium/optimium ties that to an external layer. It doesn't change your code, but it changes your risk model — the same way public vs. paid RPC changes your reliability model without changing your calls.
The short version
L2s vary on two independent axes: proof system (validity/ZK proofs = proven correct + fast finality; fraud/optimistic proofs = assumed correct + challenge window) and data availability (on Ethereum = recoverable from L1; off Ethereum = depends on an external DA layer). Cross them and you get ZK rollup, optimistic rollup, validium (ZK + off-chain DA), and optimium (optimistic + off-chain DA). "Rollup" specifically means data-on-Ethereum; move DA off-chain and it's a validium/optimium, with lower fees bought at the cost of data-withholding risk — whether you could still exit your funds if the operator disappears. It's transparent from the RPC, but it's the difference between Ethereum-grade recoverability and borrowed data availability. Know which one you're on.
Building across L2s of every flavor? A flat-rate Ethereum RPC endpoint plus rollups, validiums, and 75+ other chains under one key — HTTP and WebSocket, no per-call metering. Grab a free key and point your stack at:
https://rpc.swiftnodes.io/rpc/eth?key=YOUR_API_KEY
Originally published on the SwiftNodes blog. SwiftNodes provides flat-rate multi-chain RPC endpoints — HTTP + WebSocket, 75+ chains, no per-request metering. Grab a free key.
Top comments (0)