Every twelve seconds, a block is proposed on Ethereum. Most people read it as a heartbeat โ a steady pulse confirming that their transaction will land, their swap will settle, their liquidation will execute at a fair price. I trace the shadow before it casts.
The shadow is an auction the public never sees. Searchers push their bundles through private channels to a short list of builders. Builders compete in sealed-bid auctions for the right to construct the block. Hidden relays โ opaque intermediaries that verify bids and payments โ forward the winning block to validators who simply sign what they've been handed. A small cluster of entities now produces the overwhelming majority of Ethereum blocks. The protocol is decentralized. The market is not.
The bug hides in the beauty. Proposer-builder separation was meant to be Ethereum's elegant answer to MEV โ a structural solution so clean, so architectural, that it would neutralize the problem at its source. Instead, it created the market structure within which the problem organizes itself.
To understand why this happened, you need to walk the pipeline. MEV โ maximal extractable value โ is the rent embedded in the ability to reorder transactions. Before the Merge, this was the "dark forest": searchers competed in the public mempool, racing bots, bidding up gas, sandwiching trades in plain sight. It was chaotic, noisy โ but transparent in a strange way. At least the attacks appeared in the transaction data.
The Merge changed the game. Proof-of-stake introduced a single actor per block: the proposer, the validator whose turn it is to build. This created a bottleneck. A validator with custom tooling could capture more value, but most validators were home stakers or pools running standardized clients. They could not compete in the MEV arms race.
Enter Proposer-Builder Separation (PBS) โ concretely, MEV-Boost. The idea was elegant: proposers, who produce a block every few days, would not construct blocks themselves. Specialized builders would assemble blocks from order flow using sophisticated algorithms. Relays would shuttle blocks between builders and validators, guaranteeing payment and validity. Validators simply choose the highest-paying block. Permissionless โ anyone can be a builder. Competition would drive value back to users.
In theory, this decommodified block production. In practice, it created a market structure with three pillars: private auctions, hidden relays, and validator collusion. And the market, left to its own incentives, concentrated. The theory of "neutrality" could not survive contact with the reality of order flow economics.
The numbers trace the story. Visible MEV-Boost relay data shows a handful of builders account for the majority of all blocks. Include the private relays that do not publish their data โ the "hidden" ones โ and the real concentration is almost certainly higher. Private order flow โ transactions sent directly from wallets, aggregators, and RPC endpoints to specific builders โ bypasses the public mempool entirely. What was once a dark forest has become a gated golf course.
The problem is not cryptographic. No signature scheme was broken. No consensus vulnerability was exploited. The problem is structural, and it lives inside the trust assumptions PBS papered over.
Break down the assumptions one by one. First, the relay. A relay is expected to be a neutral messenger: it checks that the winning block is valid, confirms the builder's payment is included, and forwards the block without leaking it. But a relay also sees the entire order flow. It observes which transactions land in which blocks, which searches are running, which addresses are trading. A relay with enough market share is the settlement layer for the hidden auction. It can censor, delay, or leak. There is no protocol-level enforcement โ only a social contract that says "please don't."
Second, the builder. Builders are profit-maximizers, and their raw material is order flow. The more exclusive the order flow, the better their blocks. So they buy direct deals with wallet providers, aggregators, and protocols that route users' transactions into a builder's hands. This is the private auction: a market for order flow that ordinary users cannot see, with terms that are never disclosed. From the user's perspective, a swap is a swap. From the market's perspective, the user's transaction is a bid in an auction that happened before they knew one existed.
Third, the validators. Validators sign what the relay hands them. Their only decision is choosing the highest-paying block, and the highest-paying block comes from whoever controls the best order flow. Validators have no incentive to question the ordering, to audit whether the block contains sandwich attacks, or to ask whether the auction that produced it was fair. Every extra yield is a reward for not asking questions. When validators coordinate quietly โ sharing information, splitting streams โ everyone in the loop profits; the only loser is the user downstream.
This is what "MEV cartel" describes: not a formal organization, but a vertical structure of aligned incentives โ searchers who detect opportunities, builders who aggregate order flow, relays who control visibility, validators who rent block space. No single actor orchestrates it. The market structure aligns them into something that behaves exactly like a cartel.
There is a particular silence that hangs over a hidden relay. It accepts a transaction stream, computes a block, forwards it, disappears. No one can verify which transactions were excluded, which were deprioritized, or whether the ordering served the user or the builder's own arbitrage. Logic blooms where silence meets code โ but here, the silence is doing the heavy lifting, and the "logic" is a payment channel no auditor can inspect.
My own experience diverges from the protocols I used to analyze. In 2020, I spent months formally verifying the Curve stableswap invariant โ writing Python scripts to simulate ten thousand arbitrage attacks against its mathematical core. Back then, the battleground was math. Elegance was quantifiable. Resilience against slippage was a property you could hold in your hands and prove. This problem is not like that. There is no invariant for fairness. No formal proof for neutrality. Every attempt collides with the network effect of an extraction layer that has already absorbed the order flow.
The most uncomfortable part of the protocol-level analysis is this: the commodity traded in the private auction is not gas. It is priority over other people's transactions. In a fair market, priority is distributed by price โ the winner pays a visible, auditable cost. In the hidden auction, priority is distributed by access: who secures private order-flow deals, who learns about a transaction first, who can construct a sandwich before the victim's transaction lands. That commodity does not exist in the protocol specification. It is invisible on the consensus layer. Yet it has become the most valuable thing in Ethereum's economically active layer.
The downstream costs are quietly distributed. Ordinary DEX traders pay inflated slippage. Liquidation auctions clear at distorted prices. The AMM user base plays liquidity provider to the cartel. Finding the pulse in the static means noticing the same user pays more, every block, with zero transparency.
Now the unsettling part โ the angle that makes engineers uncomfortable. The MEV cartel is not a design failure; it is the natural output of the design. PBS did not fail to prevent cartelization. PBS created the market that cartelization optimized. Builders are not villains; validators are not corrupt. The problem is more disturbing: a protocol that defines a neutral base layer will attract a non-neutral extraction layer above it. This is as predictable as gravity.
When the community proposes enshrined PBS, inclusion lists, or auction reforms, I wonder whether we are treating the symptom. Make the auction transparent, and coordination moves elsewhere. Implement inclusion lists, and extraction shifts โ arguably to the AI agents that will soon transact autonomously. In 2025, I co-authored a security framework for AI agents transacting on-chain; the attack surface was hallucination, not fraud. Extend that to the MEV layer: autonomous agents producing predictable order flow, builders optimizing it. The cartel will not be dismantled by a better mechanism. It will be reformed around it.
The deeper blind spot is that the cartel's power does not come from validators. It comes from order flow origination. Whoever controls the user's path to the chain controls the auction. Wallets, RPC providers, aggregators โ these are the unexamined gatekeepers. Add a regulator, and the picture sharpens. In traditional markets, private auctions where insiders see and act on other people's order flow are called market manipulation. The CFTC's playbook against spoofing and front-running maps onto the cartel's behavior โ except the blockchain version has no compliance layer to subpoena. Hidden relays are dark pools no regulator has visited. Yet.
Vulnerability is just a question unasked. The question is not whether the MEV cartel exists โ it does, and its fingerprints are on every block. The question is whether we accept a block market that was never neutral, or design the next one better. Logic blooms where silence meets code, but only if we force the silence to speak. The first step is measurement: builder concentration, private relay share, excluded transactions. The second is harder โ admitting that every DeFi user trading without MEV protection sponsors a system they cannot see. In the void, the bytes whisper truth. We just have to listen.