Level 3 · How it worksThe village#260

Data availability explained: a summary you can't check is a promise, not a record

Data availability means the data behind a block or a rollup batch has really been published, so anyone can download it and check it. Without it, a rollup's summary can't be challenged and its users can't work out their own balances.

Ben has offered to keep the market's tally. All day he notes who sold what to whom. In the evening he writes one line into the village notebook: "Everyone's balances, all correct."

The tally sheet itself? He keeps it at home. It's very tidy, he says. No need to look.

Anna says there is every need to look. If nobody can read the sheet, nobody can spot a wrong line, and nobody can prove what they own if Ben disappears. A summary you can't check is a promise, not a record. That is the whole idea of data availability.

What it is

Data availability means the data behind a block, or behind a rollup's batch, has actually been published, so that anyone who wants it can download it and check it. It's not about keeping data forever. It's about making sure the pages existed and could be read when it mattered.

ethereum.org describes the problem as "the need to prove to the whole network that the summarized form of some transaction data" really stands for data that was published (ethereum.org). The hard part is proving it without making every node download everything.

Why rollups can't skip it

A rollup posts summaries of its users' transactions to Ethereum. That only works if the data under the summary is out in the open.

  • Optimistic rollups rely on someone spotting a bad batch and submitting a fraud proof. You can't prove fraud in a sheet you can't read.
  • ZK-rollups post a proof that the sums are right. But a proof doesn't tell you your balance. Without the data, users can't rebuild the state and leave if the operator stops.

The authors of EIP-4844 put the requirement precisely: "Rollups need data to be available once, long enough to ensure honest actors can construct the rollup state, but not forever" (EIP-4844).

How it works: checking without downloading everything

The trick has two parts, set out in a 2018 paper by Mustafa Al-Bassam, Alberto Sonnino and Vitalik Buterin (arXiv).

Part one: stretch the data. Before publishing, the block producer extends the data with extra, redundant pieces (this is called erasure coding). Ethereum's version means "any portion of the data can be reconstructed from any existing 50% of the whole" (ethereum.org). So hiding one line is useless. To make anything truly unrecoverable, Ben has to hide more than half of the stretched sheet.

Part two: ask for random pieces. Each checker requests a few random pieces. If more than half is hidden, each request has at least a one-in-two chance of landing on a gap. The chance that every request succeeds anyway halves with each one:

  1. 10 samples: at most 1 in 1,024 that Ben gets away with it.
  2. 30 samples: at most about 1 in 1.07 billion.
  3. 100 samples: about 1 in 10 to the power of 30. ethereum.org gives the same figure: "for 100 chunks the chance is 10^-30".

Each checker downloads a handful of pieces, not the whole sheet, and still gets near-certainty. That is data availability sampling. Good luck, Ben.

Where it runs. Ethereum's Fusaka upgrade of 3 December 2025 added PeerDAS, a version of this for rollup data: the data is split into columns across 128 subnets, and each full node holds "only 1/8th of the data" (ethereum.org). Celestia uses a two-dimensional version, extending a square of data to twice the width and height, and its light nodes sample random squares from it (Celestia docs).

Why it matters

  • Who gains. Rollup users. If the data is out there, anyone can check the operator's work and rebuild their balance without the operator's help.
  • Who pays. Rollups, for the space. Every byte they publish on the main chain is paid for, which is why cheaper data space, covered in "Blobs and danksharding explained", matters to them and their users.
  • A market of keepers. Some projects offer data availability as a separate service. Celestia describes itself as a layer that "orders blobs and keeps them available while execution and settlement live on layers above" (Celestia docs). EigenDA is "built on EigenLayer" and secured by restaked ETH (EigenCloud docs). Examples, not recommendations.
  • Where the risk moves. A rollup that posts its data somewhere other than Ethereum is only as safe as that other keeper. Cheaper pages, different guards.

Where people get it wrong

  • "Data availability means storing data forever." It means publishing it so it can be checked. Ethereum's blob data can be deleted after about 18 days (EIP-4844); keeping history is a separate job.
  • "A zero-knowledge proof makes the data unnecessary." The proof shows the sums are right. You still need the data to know what you own and to leave without the operator.
  • "Sampling means trusting a few nodes." Each checker samples on its own, at random. A cheat would have to guess everyone's picks in advance.
  • "All data availability layers are as safe as Ethereum." Each has its own set of guards. Who they are, and what they lose if they lie, is the question to ask.

Ask what the keepers lose

"Rollups explained" tells you to find where a rollup posts its data. Go one step further. If the answer is anywhere other than Ethereum itself, find out three things: how many keepers hold the data, who can check them by sampling, and what a keeper loses if it withholds a page.

If the answer is "a few named companies, nobody samples, nothing", you have found Ben's tally sheet. Tidy, he says. No need to look.

Frequently asked questions

What is data availability in blockchain?

Data availability means the data behind a block or a rollup batch has been published, so anyone can download and check it. It matters because a summary whose underlying data is hidden can't be challenged, and users can't rebuild their balances from it.

Why do rollups need data availability?

Optimistic rollups depend on someone checking batches and submitting fraud proofs, which needs the data. ZK-rollups prove their batches are correct, but users still need the data to know their balances and to withdraw if the operator stops. EIP-4844 says rollups need data available long enough for honest actors to rebuild the rollup state.

What is data availability sampling?

It is a way to check that data was published without downloading all of it. The data is first extended with redundant pieces, so that hiding any of it means hiding more than half. Each checker then requests random pieces, and with 100 samples the chance of being fooled is about 1 in 10 to the power of 30.

What is PeerDAS?

PeerDAS is the data availability sampling system Ethereum added with its Fusaka upgrade on 3 December 2025. It splits rollup data into columns spread across 128 subnets, so each full node holds only about an eighth of the data instead of all of it.

What is a data availability layer?

It is a network that specialises in publishing data and keeping it available for others to check, while transactions are executed elsewhere. Celestia and EigenDA are two examples, not recommendations. A rollup that uses one depends on that layer's security instead of Ethereum's for its data.

Sources

  1. ethereum.org, Data availability (page updated 22 October 2026)
  2. Mustafa Al-Bassam, Alberto Sonnino, Vitalik Buterin, Fraud and Data Availability Proofs (arXiv, 2018)
  3. EIP-4844, Shard Blob Transactions
  4. ethereum.org, Fusaka (PeerDAS) (page updated 16 October 2026)
  5. Celestia docs, Data availability layer
  6. EigenCloud docs, EigenDA overview

Read next.