data availability sampling

Data availability sampling (DAS) lets a lightweight node become highly confident that a large block of data was fully published without downloading all of it. The intuition is statistical: if you want to be nearly certain a giant lottery drum is not missing any balls, you do not count every ball — you reach in blindly a few dozen times, and if you never grab an empty slot, you conclude with overwhelming probability that the drum is full. DAS does the same to data, and it is the technique that makes Ethereum's danksharding roadmap feasible because it breaks the rule that securing data requires downloading it.

The cryptographic trick that makes sampling sound is erasure coding. Before publishing, the data is expanded with a Reed-Solomon code so that the original can be reconstructed from any sufficiently large fraction of the pieces — for example, any 50% of the extended data suffices. This changes the adversary's problem dramatically: to make the data unrecoverable, a malicious block producer must now hide more than half of all the pieces, not just a tiny sliver. A sampler who requests a few random pieces and receives them all has a vanishing probability of missing such large-scale withholding; each successful sample roughly halves the chance that the producer is hiding enough to make reconstruction impossible.

In practice each light node privately requests a handful of random chunks (with KZG commitments proving each chunk belongs to the committed data); once enough independent nodes have each sampled successfully, the network collectively guarantees the whole block is recoverable even though no single node downloaded it all. Ethereum's full danksharding adds a second dimension (2D erasure coding) so that missing pieces can be reconstructed row-by-row and column-by-column. DAS is what allows a blockchain to safely offer far more data throughput than any individual node could store — the cornerstone of cheap rollup data.