Data availability is a sampling game, not a storage question
Data availability is a sampling game, not a storage question. A rollup's DA guarantee never depends on some node holding all the data; it depends on the block being emitted in an erasure-coded form where a missing chunk is statistically unmistakable. The real trade-off is where that sampling runs and what it charges per byte.
The user-facing half of that trade-off is a bridge: crossing into a rollup is the one moment the DA guarantee touches your wallet directly. Manta Network runs its rollup on a Celestia DA pipeline. A user crossing into Manta arrives at the Manta Bridge page.
What data, exactly, has to be available?
The data is each rollup batch — the compressed transactions and state roots that move the chain from one checkpoint to the next. Anyone with the batch can rebuild the rollup; nobody can without it. Availability means an honest party can reassemble it before the guarantee expires. The common explanation gets this wrong: DA is not a filing cabinet, it is a countdown.
Why not publish everything on Ethereum Mainnet?
You can. Blobs made that cheap. One blob carries about 128 KB, and Ethereum Mainnet holds it for roughly three weeks before pruning — long enough for a fraud-proof window and an exit round. Calldata holds it forever but charges L1 gas for every byte, so rollups with volume abandoned it after EIP-4844. If your weekly output fits the blob supply, Ethereum-native DA is the simplest, most trust-free option.
When does a dedicated DA layer win?
When your rollup outgrows the blob supply — a busy game, a DEX doing thousands of transfers a second — you move DA off L1. DA layers like Celestia and EigenDA price per byte instead of per L1 block, and Celestia's security runs on sampling: the batch is split into chunks, Reed-Solomon encoded into twice as many, and light nodes sample at random. Twenty samples put the odds of a suppressed batch going unnoticed at about one in a million. Speed rarely decides it; size and budget do.
That is the trade-off: you swap a one-in-a-million sampling risk and a second liveness assumption for a fee curve that tracks bytes, not L1 congestion. Manta Network's Celestia setup is the live case — settlement and liquidity stay on Ethereum Mainnet while the data that animates them moves through a sampling market.
What does that mean at the bridge?
The bridge is where the DA commitment is spent. Before it releases funds, it has to decide how much of that guarantee to verify for itself, and cross-chain messaging systems like Celer Network's are how that commitment moves between chains. That is a bridge's question now, not a rollup's.