Celestia light nodes verify that a block's data was published without downloading it

A Different Question at the Heart of Celestia

Most blockchain light clients rely on full nodes to tell them whether block data has been published. @Celestia takes a different approach. Data availability answers one question: has this data been published? A node verifies this when it receives a new block being added to the chain. Rather than trusting fuller nodes at their word, Celestia gives light nodes a way to check for themselves.

Data availability sampling (DAS) is a process where light nodes request a random sample of block data, allowing them to verify that the block data is available without downloading the entire block. That distinction matters. While full nodes can verify a block by fully downloading it, light clients traditionally have no such mechanism since they only download the block header. DAS provides a method to verify data availability without downloading the entire block, letting them retain minimal resource requirements.

How the Sampling Mechanism Works

The process rests on a specific encoding scheme. Block data is split into k x k chunks arranged in a matrix and expanded with parity data into a 2k x 2k Reed-Solomon encoding matrix, producing independent Merkle roots calculated from rows and columns. These Merkle roots form the block data commitments stored in the block header.

Every light node randomly chooses a set of unique coordinates in the extended matrix and queries bridge nodes for the data shares and the corresponding Merkle proofs at those coordinates. Light nodes conduct multiple rounds of this random sampling, with each completed round increasing confidence that data is available. Once a node reaches a predetermined confidence level, such as 99%, it considers the block data available.

A block producer attempting to withhold data faces a structural problem. Every received data share with a correct Merkle proof is gossiped to the network. Enough unique samples across honest nodes means the full block can be reconstructed, making a withholding attack self-defeating.

The scalability implication runs counter to the usual logic. Data availability sampling enables light nodes to contribute to both the security and throughput of the network with significantly cheaper hardware than that of full nodes. More light nodes participating in sampling collectively supports larger blocks rather than adding load.

There are assumptions to keep in mind. Light nodes performing DAS must have a guarantee that the sampled data is erasure coded correctly. In Celestia, light nodes can be notified of a maliciously encoded block through Bad Encoding Fraud Proofs (BEFPs). That mechanism depends on connectivity. Light nodes must be connected to at least one honest full node to receive fraud proofs for incorrectly erasure coded blocks. If a light node is not connected to an honest full node, it cannot verify that the block has been improperly constructed. Eclipse attacks, which cut a node off from honest peers, can break that guarantee.

Sources:
Celestia Docs: Modular Data Availability Layer
Celestia Glossary: Data Availability Sampling
L2BEAT: Celestia Data Availability