Overview

  • Have a set of validators, who have staked >= 32 ETH into a deposit contract.
  • On a fixed schedule (12s slots, 32 slots in an epoch), a validator collections transactions and proposes a block.
    • This is the “beacon block” that contains a proposed ETH block, plus information about: rewards, penalties, slashings, attestations etc.
  • Randomly, a set of validators are chosen to verify/validate the results of the proposed block.
    • Validators “attest” that the block is both 1. valid, 2. the logical next block/state of the chain (i.e. resolve forks ).
  • Transactions are confirmed when:
    • Two checkpoints, start of each epoch
    • “Supermajority link” between two checkpoints, i.e. both checkpoints must have attestations from >2/3 of staked ETH.
  • Various details (i.e. who are validators, proposed blocks, withdrawal accounts, etc) are stored on chain.

Staking Economics

  • Validators get ETH for honestly proposing and validating blocks.
  • 32 ETH are collatoral for dishonest behaviour (and are “slashed”, or lose some fraction of their staked ETH, accordingly). Namely
    • Equivocating: Propose >1 block per slot
    • Contradictory attestations: I.e. beacon block is valid, but attesting to it being invalid.
  • Badly behaving validators will be slashed, and eventually removed from the network after 36 days.
    • On day 18, they receive a dynamic penality, a “correlation penalty”, depending on how correlated their slasheable activities correlated with the slashing volume on the network.
  • The economic incentives of staked ethereum attempt to make attacks expensive.

Details

Forks

  • POS can still have forks. For example, from standard distributed systems unreliability (e.g. network delays), or equivocating.
  • Fork resolution via LMD GHOST, essentially looking for fork with greatest historical attestations.
    • Interestingly, this means attempts at 51% attacks may be costly, but can still win. Honest minority validators could split off and try to attract back apps/exchanges/etc. With support, they could then, again, make the 51% attackers lose their staked ETH.