Data availability: the quiet constraint behind every rollup

Data availability is the guarantee that the information needed to verify a chain's state is actually published where anyone can get it. It matters because verification is worthless without it: if a rollup operator can compute a new state but withhold the transactions that produced it, nobody can prove fraud, exit independently, or reconstruct their balance. Availability is what turns a claim into something checkable.

Why is availability a separate problem?

Three things must hold for a scaling system to be trustworthy: the state transition must be valid, someone must be able to check that, and the inputs must be obtainable. Validity proofs address the first two. They say nothing about whether the data exists anywhere you can reach — and a proof you cannot audit against inputs you cannot see leaves you dependent on the operator.

The concrete failure is the data withholding attack. An operator publishes a state root, keeps the transaction data private, and users cannot determine their own balances or construct the proof needed to challenge anything. The chain looks fine and nobody can act.

How do rollups solve it?

A rollup posts its transaction data to a base chain. Anyone can download it, replay the transactions, and reconstruct the rollup's state independently — which is what allows a user to exit through the canonical bridge even if every operator disappears. That is the property that distinguishes a rollup from a chain that merely settles somewhere.

The cost, historically, was severe: base-layer space was priced for permanent storage, and rollups were the largest consumers of it. Our guide to Ethereum rollups covers how the batching works in practice.

What changed with blobs?

Dedicated blob space introduced a data type that is cheap, large and deliberately temporary: consensus nodes keep it long enough for anyone to download and verify, then discard it. Since availability only needs to hold long enough for verification and challenge, permanent storage was never required — the cost had simply been an artefact of the original design.

What is off-chain data availability?

Some systems post only the proof to the base chain and keep the data with a committee or a separate network. This is much cheaper and changes the guarantee: users now depend on that committee publishing when asked. A system doing this is generally called a validium, and the distinction from a rollup is exactly this one — our guide to sidechains versus rollups sets out the family tree.

How does sampling work?

Downloading everything to check availability defeats the purpose of light verification. Data availability sampling has clients request many small random pieces of the data; if the pieces come back, the probability that a meaningful portion is missing falls sharply. Combined with erasure coding — which lets the whole be reconstructed from a subset — a small number of successful samples gives high confidence that the full data exists.

Why should a user care?

• It determines whether you can exit a layer 2 without the operator's cooperation.
• It explains why fees on rollups fell so sharply once dedicated data space existed.
• It is the single clearest question separating designs that market themselves with the same words.
• It tells you what happens on the worst day, which is the only day the guarantee matters.

When you evaluate a network, ask where its data goes and who can withhold it. That one question distinguishes systems that inherit a base chain's security from systems that merely borrow its name — see our guide to layer 1 versus layer 2 for the wider framing.

How do you check where a network's data goes?

This is documented, not secret, and the answer is usually one page deep in a project's technical docs. Three checks settle it.

• Look for the term used to describe the design — rollup, validium, optimium, sidechain — and then verify it against the data question rather than accepting the label.
• Find where batches are posted. If they land on the base chain, look for the contract that receives them; if they go to a committee or a separate network, find out who sits on it.
• Check what a forced exit looks like: a rollup documents how to withdraw using only base-chain transactions, without the operator's cooperation.

Independent trackers publish comparisons of these properties across networks, which is a faster starting point than reading each project's own description. Whichever source you use, the question to answer is the same: on the day the operator stops helping you, can you still prove what you own?