Vitalik Buterin (@vitalik.eth): [Farcast] @vitalik.eth In the first version, there are two cases where the full data of a block still needs to exist in one place: (i) initial broadca…

[Farcast] @vitalik.eth In the first version, there are two cases where the full data of a block still needs to exist in one place: (i) initial broadcasting, (ii) reconstruction, in case a publisher publishes 50% <= p < 100% of a block. But these roles are untrusted: we only need one honest actor to do them, if there are also 100 dishonest actors the protocol simply bypasses them. And different nodes can perform this task for different blocks. In the future, cell-level messaging and distributed block building will allow even these two functions to be distributed. This is all new technology, and the core devs are wise to be super cautious on testing, even after they have been working on this for years. This is also why the blob count will increase conservatively at first, and then become more aggressive over time. But it is the key to L2 scaling (and eventually L1 scaling, once the L1 gas limit goes high enough that we have to put L1 exec data into blobs)