Erasure Coding
Erasure coding is a method of adding mathematically calculated redundancy to data so the original can be rebuilt even if some pieces are lost or withheld. Reed-Solomon coding is the most widely used type.
Erasure coding splits data into chunks and then generates extra chunks from them. In Reed-Solomon coding, the original chunks are treated as points on a polynomial, and further points on the same polynomial become the extra chunks. If data is split into k pieces and extended to n pieces, any k of the n are enough to recover everything. The technique long predates crypto and is used in QR codes, CDs and data storage systems.
In blockchains it matters mainly for data availability. Rollups and sharded designs need assurance that block data was actually published, not hidden by the block producer. If a block's data is doubled with erasure coding, a producer would have to withhold more than half the pieces to stop it being rebuilt. That makes hiding data much easier to detect. Light nodes can download a few random pieces, a technique called data availability sampling. If every piece they request is available, they can be highly confident the full data is available too. Decentralized storage networks use the same idea to keep files recoverable when some nodes go offline.
For example, 4 MB of data might be split into four 1 MB chunks and extended to eight. Any four of the eight, in any combination, reconstruct the original.
A common misconception is that erasure coding protects privacy or proves data is valid. It does neither. The chunks are not encrypted, and a dishonest producer could encode the data incorrectly, so systems pair it with fraud proofs or cryptographic commitments to catch bad encoding.