A block can be valid in form while its data is withheld, which prevents anyone from checking the transactions inside. Data availability sampling is a method for detecting that without downloading everything.
Withheld data is a distinct failure
Fraud proofs let honest participants demonstrate that a block contains an invalid transition, but constructing one requires access to the block's contents.
A producer who publishes a header while withholding data blocks that process entirely. Nothing can be proven wrong because nothing can be examined.
Downloading every block defeats the purpose of light verification, so the problem is confirming publication without paying the cost of full retrieval.
Erasure coding changes the odds
Before publication, block data is expanded with redundancy so the original can be reconstructed from any sufficiently large subset of the pieces.
This means withholding a small fraction is useless, since the rest is enough to rebuild it. Effective withholding requires hiding a substantial portion.
Hiding a substantial portion is exactly what random sampling detects, because a random request is then likely to land on a missing piece.
Many small checks add up
Each light node requests a handful of randomly chosen pieces. If all arrive, that node gains modest confidence the data is available.
With many independent nodes sampling different pieces, the combined coverage becomes strong even though no participant retrieved much.
The security therefore emerges from the population rather than from any individual, which is why participation levels matter to the guarantee.
Commitments prevent substitution
Sampling would be meaningless if a producer could return arbitrary bytes, so responses are checked against a commitment published in the header.
Polynomial commitment schemes let a node verify that a returned piece belongs to the committed data without holding the rest of it.
Incorrect encoding is a separate concern, handled either by proofs demonstrating the coding was done wrong or by schemes that make incorrect encoding impossible to commit to.
Why rollups depend on this
A rollup executes transactions elsewhere and posts data to a base layer so anyone can reconstruct its state independently.
If that data is unavailable, users cannot compute their own balances or exit without the operator's cooperation, which removes the property that distinguishes a rollup from a private ledger.
Dedicated availability layers make publishing that data cheaper by specializing in exactly this guarantee, rather than also executing the transactions themselves.
The separation is why availability is now discussed as a distinct service with its own pricing. A rollup chooses where to publish, and that choice determines who must be trusted for its users to reconstruct their own balances.