Proof of stake and the beacon chain

Time on the consensus layer is divided into slots of twelve seconds, grouped into epochs of thirty-two slots. In each slot one validator is chosen to propose a block, and a committee of validators is chosen to attest to the block they see. If the chosen proposer is offline, the slot passes without a block; the chain does not stall, it simply misses that slot. That is a meaningful difference from proof of work, where a missed block is not a concept because blocks are produced whenever a miner finds one.

Committees exist to keep the signature load manageable. If every validator attested to every block, the number of signatures per slot would grow with the size of the validator set and the chain would become unusable. Instead the active set is partitioned into committees, each committee attests in its assigned slot, and the attestations within a committee are aggregated into a single signature before they are included in a block. The committee size is a parameter chosen so that a committee is large enough that a randomly selected majority is implausible, and small enough that the aggregation is cheap.

The rotation of committee membership is what makes the assignment unpredictable. A validator cannot know far in advance which slot it will attest in, and the assignment is derived from the chain's own randomness, so an attacker cannot position itself to control a committee without controlling a large share of the whole set. The trade-off is that a validator must be online for its assigned slots or forgo the reward, which is why running a validator is an availability commitment rather than a passive one.