The validator that wins a slot is supposed to build a block. In practice almost none of them do: they take a bid from a specialist builder and sign what arrives. The gap between the protocol's description of itself and what actually happens is what the next round of changes is about, and the reason is less ideological than it looks.
What enshrining the split changes
The separation already exists through software sitting outside the consensus rules. EIP-7732 moves it inside, scheduled for the Glamsterdam hardfork. The slot is redrawn around it: the attestation committee votes on the beacon block, the builder publishes the execution payload a few seconds later having watched the attestation weight accumulate, and a dedicated committee then votes on whether the payload was actually available.
That last committee has 512 members against roughly 30,000 attesters in the ordinary one. It exists to answer a single narrow question — did the payload show up in time — and is deliberately small enough to answer it quickly.
The consequence for builders is that they become staked participants, depositing a minimum of 1 ETH. That makes trustless payment possible: the consensus layer can debit a builder's stake and credit the proposer directly, instead of relying on an execution-layer transaction the way the current arrangement does.
Why the payment path matters so much
Because of how the money is distributed. Data for the first half of 2026 puts the top 0.1% of slots at roughly a quarter of all extractable value, concentrated in blocks worth hundreds of ETH. Almost every slot is worth very little and a handful are worth a great deal.
A system where value is spread evenly can tolerate a sloppy payment path, because nobody is strongly tempted. A system where one slot in a thousand carries a quarter of the money cannot, because that is precisely the slot where somebody will test what happens if the payload never arrives.
The timing game that is already running
A proposer can hold its block until close to the attestation deadline. Honest validators, seeing nothing new, attest to the parent; the delayed block then becomes canonical, and those attesters forfeit their head-vote reward. The measured loss is an average 15.06% of total reward — not an occasional accident but a persistent transfer from validators behaving correctly to one choosing when to broadcast.
The proposed remedy is narrow by design: a distance-weighted head reward, where an attestation pointing at a recent canonical ancestor earns a partial reward rather than nothing, while voting the freshest canonical head still earns the maximum. That puts the loss at 4.71%. What it deliberately leaves alone matters as much — fork choice and proposer boost are untouched, so the change is to the reward calculation rather than to consensus, which keeps the blast radius small.
The asymmetry is the point rather than the percentage. Any reward rule that penalises honest behaviour because of an adversary's choice is a slow incentive to stop behaving honestly.
What this means for anyone building on top
| Question | Today | Under EIP-7732 |
|---|---|---|
| Who constructs the block? | A specialist builder, via software outside consensus | The same, with the arrangement written into the protocol |
| How is the proposer paid? | An execution-layer transaction from the builder | The consensus layer debits a staked builder directly |
| Who confirms the payload arrived? | Nobody, as a protocol matter | A 512-member payload timeliness committee |
| What stops a late broadcast? | Nothing; honest attesters absorb the loss | A proposed distance-weighted reward, still under discussion |
For a protocol designer the practical reading is that block space is an auction with a very heavy tail, and the mechanisms around it are being hardened at the points where that tail makes misbehaviour worth the effort. For everyone else it is a reminder that the neutrality of inclusion is a property somebody maintains, not a property the network has by nature.
Questions
- What is proposer-builder separation?
- An arrangement in which the validator that wins a slot does not construct the block but takes a bid from a specialist builder. It already happens through software outside the consensus rules; EIP-7732 writes it into the protocol, scheduled for the Glamsterdam hardfork.
- How concentrated is the value in Ethereum blocks?
- Extremely. Data for the first half of 2026 puts the top 0.1% of slots at roughly a quarter of all extractable value, in blocks worth hundreds of ETH, while almost every other slot is worth very little.
- What is the payload timeliness committee?
- A committee of 512 validators that votes on a single question — whether the execution payload was published in time. It is deliberately small against the roughly 30,000 ordinary attesters so that it can answer quickly.
- What does a late block cost honest validators?
- An average 15.06% of total reward, because attesters that voted for the parent forfeit their head-vote reward when the delayed block becomes canonical. A proposed distance-weighted reward would reduce that to 4.71% without touching fork choice.