How Ethereum Adds a Block

Where this sits

• Ethereum is a single shared computer that thousands of machines run in lockstep. Adding a block is how it takes one step forward.
• This lesson covers only that step: who proposes a block, who votes on it, and when the result stops being reversible.
• Nothing here assumes prior Ethereum knowledge. Later lessons assume this one.

The clock

• Time is cut into slots of 12 seconds. Thirty-two slots make an epoch, so an epoch is 6.4 minutes.
• Every slot, one validator is picked at random to propose a block. Selection uses RANDAO, an on-chain randomness value each proposer contributes to.
• Every slot, a randomly chosen committee of validators attests — publishes a signed vote about what it sees.
• Committees are sized so that every active validator attests exactly once per epoch, not once per slot. That keeps the message volume survivable.

The clock runs whether or not anyone proposes. A slot with no block is an empty slot; the chain continues at the next one.

Validators

• A validator is a stake plus a signing key, not a machine. One machine can run many.
• 32 ETH is the minimum to activate — MIN_ACTIVATION_BALANCE.
• Since EIP-7251, a single validator can hold up to 2048 ETH of effective balance. Before it, everything above 32 ETH was idle and operators ran thousands of separate validators instead.
• Stake is collateral. Voting for two conflicting things is slashable: provably contradictory signatures cost the validator part of its stake and eject it.
• Being offline is not slashing. It leaks a small amount, roughly the size of what would have been earned.

What an attestation says

One attestation carries three claims at once, and the two halves of consensus read different parts of it.

A checkpoint is just the block at the first slot of an epoch. Voting happens on blocks every slot, but finality is decided on checkpoints once an epoch — which is why finality moves in 6.4-minute steps rather than 12-second ones.

Fork choice: which chain is real right now

• Two proposers can produce blocks that conflict — through latency, a network split, or on purpose. Nodes need a rule that picks the same winner without asking each other.
• The rule is LMD-GHOST: follow the fork with the greatest weight of attestations behind it. Weight is staked ETH, not validator count.
• "LMD" means latest message driven — only each validator's most recent vote counts, so an old vote cannot be replayed to swing the head.
• Fork choice is an opinion that updates every slot. It can change. A block that was the head can stop being the head — a reorg.

Finality: which chain can never change

• Casper FFG runs on top of fork choice and produces a stronger guarantee: not "this is the head" but "this can never be undone".
• Validators vote for pairs of checkpoints, source and target. When a pair collects attestations representing at least two thirds of all staked ETH, that is a supermajority link.
• The newer checkpoint of the pair becomes justified. When a justified checkpoint gets a supermajority link of its own, the earlier one becomes finalized.
• Practically: a block is usually finalized about two epochs after it lands, roughly 13 minutes.

Why it matters

There are three different meanings of "confirmed"

• Anyone crediting value on the basis of an on-chain event — exchanges, bridges, payment processors — is choosing a point on this table, whether they know it or not.
• Bridges are where the choice gets expensive. Acting on a head-of-chain event on one chain and settling on another leaves no way to undo the second half if the first reorgs.
• The reason to wait is not that the network is slow. It is that fork choice is designed to change its mind and finality is designed not to.

Liveness and safety are separate failures

• If more than a third of stake goes offline, no supermajority link forms and finality stalls while blocks keep being produced. The chain works; it just stops making promises.
• An inactivity leak then drains the offline validators' stake until the online set is two thirds again and finality resumes.
• This has happened on testnets — including on Glamsterdam's Platåberget in August 2026 — and it is a recovery path, not a crash.

The proposer has a monopoly for 12 seconds

• Whoever proposes a slot chooses which transactions go in and in what order. That power is worth money, which is what MEV is.
• Nearly every proposer therefore outsources block building to specialists rather than building blocks itself. That arrangement, and the protocol changes to formalise it, are their own topic.

Where to go deeper

• ethereum.org's proof-of-stake docs — the same ground with the attestation and rewards detail this lesson skipped.
• EIP-7251 for what raising the effective balance cap changed for operators.
• A block explorer's slot view (beaconcha.in, Dora): watch a live epoch justify and then finalize. It makes the 6.4-minute rhythm concrete faster than any description.