Producing a block used to be one job: a validator selected to propose the next block would also decide which transactions to include and in what order, then broadcast the result. That combined role is exactly what makes MEV extraction possible — whoever builds the block can also profit from how they order it. Proposer-builder separation splits that single job into two.

A builder's job is purely construction: assemble transactions from the public mempool and private order flow into the most profitable possible block, competing against other builders doing the same thing. A proposer's job is purely selection: look at the blocks builders have offered and pick the most profitable one, without needing to know or care what's actually inside it.

This division does something specific to the MEV problem: it turns block construction into a competitive market rather than a private advantage. A proposer who used to have to be sophisticated enough to extract MEV themselves now just picks the highest bid from specialists competing against each other, and that competition compresses the specialised advantage down toward a fee the proposer receives regardless of their own technical capability.

The mechanism that lets a proposer commit to a block without seeing its contents first is the relay. A builder submits its block to a relay along with a bid; the relay verifies the block is valid and forwards only the bid and a commitment to the proposer, who signs a commitment to that specific block sight-unseen, trusting the relay to reveal the real contents only after that commitment is made. This is what stops a proposer from simply stealing a builder's profitable transaction ordering for themselves.

That trust requirement is also the design's obvious weak point. A small number of relays currently handle the overwhelming majority of this traffic, and a relay that misbehaves, censors specific transactions, or goes offline can distort block production for every proposer that depends on it — recreating, at the relay layer, a version of the centralisation problem the split was partly meant to solve at the proposer layer.

The direction of ongoing work is toward removing that dependency entirely: enshrining a version of this separation directly into the base protocol rather than relying on an off-chain market of relays, so the guarantee doesn't rest on a small number of intermediaries behaving well.