Data Availability Sampling

Data Availability Sampling solves a specific trust problem: how can a node with limited storage and bandwidth be confident that all of a block's data was actually published, not just its header? Without an answer, a dishonest block producer could withhold part of a block while still claiming it is valid, and light clients would have no way to catch the fraud short of downloading everything themselves.

DAS gets around this using erasure coding. Before publishing, block data is mathematically expanded with redundant "parity" pieces, so the original content can be fully reconstructed even if a large chunk of the expanded data goes missing. A node then requests a handful of small, randomly chosen pieces instead of the whole block. If the data was genuinely withheld, some of those random requests will fail, and the probability of missing that failure shrinks rapidly with each additional successful sample, reaching near-certainty after only a few dozen checks. Cryptographic commitments, such as KZG proofs, let a node verify that each sampled piece genuinely matches the original data.

This matters because it lets lightweight nodes help secure a network without downloading megabytes of data per block, which is what makes scaling to many rollups practical. On Ethereum, DAS underpins the EIP-4844 blob mechanism and its successor, PeerDAS, part of Ethereum's broader sharding roadmap that eventually aims for full danksharding. Dedicated data-availability networks like Celestia and Avail use similar sampling techniques as their core product, letting Layer 2 rollups post data more cheaply while still inheriting strong guarantees that it was actually published.