Part 3 of the airgap-closure series. Following on from Part 1 (the airgap problem in DeFi) and Part 2 (augmented mechanism design as methodology), this post argues that the MEV abstraction is too narrow, and the right unit of analysis is generalized extractable value. Extraction is conserved across protocol layers, which means partial fixes relocate rather than eliminate.
The narrow frame
MEV discourse since Daian et al. has converged on transaction-ordering extraction as the central problem. Flashbots, MEV-Share, proposer-builder separation, encrypted mempools, and CoW-style solver auctions are all responses to that frame. They share a common premise: that the disease is ordering, and that addressing ordering is sufficient.
The premise is too narrow. Ordering is one channel through which value flows asymmetrically from participants who generate it to participants positioned to capture it. There are at least six others, and any system that eliminates ordering-based extraction without addressing the rest has not eliminated extraction, it has relocated extraction to the unaddressed channels.
What is conserved
Every participant in a protocol generates some quantum of value through their participation. Liquidity providers supply capital that absorbs price volatility. Traders supply order flow that creates price discovery. Oracle operators supply data. Borrowers supply collateral. Validators supply consensus security. The protocol distributes the resulting joint value back to participants under some allocation rule.
Define the structural extraction of a protocol as the total value received by participants in excess of their Shapley value (their marginal contribution to the joint game averaged across all coalitions). When this quantity is positive, some participants are receiving more than their marginal contribution warrants, which means others are receiving less. The asymmetry exists not because anyone is acting badly but because the protocol’s allocation rule embeds a structural channel through which value flows disproportionately.
Generalized extractable value is this quantity. MEV is one channel. There are at least six others, and the conservation property says they cannot be eliminated by patching any single channel in isolation.
The seven channels
Transaction-ordering extraction is what the literature calls MEV. Mempool visibility plus sequential execution creates the structural condition. Builders, searchers, and validators positioned at the ordering chokepoint capture value from traders whose orders are visible before execution. Sandwich attacks, front-running, and just-in-time liquidity provision are the operational expressions.
Governance extraction is value captured by concentrated token holders or bribe markets from users subject to governance parameters. The structural condition is that token-weighted voting allocates control over economically consequential parameters to whoever holds tokens, which under typical distributions concentrates control among early holders, large allocators, and the secondary markets that bid for voting rights (Convex, Votium, and analogues).
Token rent-seeking is value captured by token holders earning fees from mandatory token intermediation. The structural condition is that the protocol requires its native token to access functionality (staking, fee discounts, governance participation), creating ongoing rent that users pay for access to functionality that does not strictly require token mediation.
Capital-formation extraction is the asymmetry between early investors with pre-public allocations and public-market buyers who acquire the same token later. The structural condition is the standard pre-public sale to insiders followed by listing, where the price discovery from the listing creates value that accrues to pre-public buyers.
Oracle extraction is value captured by operators of the off-chain to on-chain data pipeline. The structural condition is that on-chain protocols depend on off-chain data, and whoever controls the pipeline can charge for the privilege of supplying that data, or arbitrage latency between the off-chain price and the on-chain reference price.
Platform extraction is value captured by platform operators or data brokers from users whose activity generates the data being captured. The structural condition is that user activity on a platform produces logs, profiles, and behavioral data that the platform controls, monetizes, and uses to improve its position relative to users.
Liquidation extraction is value captured by liquidation bots and sequencer operators from borrowers whose positions become liquidatable. The structural condition is that liquidation ordering determines who captures the liquidation discount, and sequencer-privileged ordering or first-mover bot networks capture it systematically.
These seven exhaust the categories I have been able to identify across major DeFi protocols. There may be others. The argument does not depend on the count being exactly seven; it depends on the count being greater than one.
Why conservation matters
If a protocol addresses transaction-ordering extraction alone, governance extraction remains available. Governance token holders extract rent through parameter manipulation even when ordering attacks are impossible. The total extractable value is unchanged; the extraction has relocated.
If a protocol addresses ordering and governance but retains a rent-seeking native token, token-rent extraction remains. If a protocol eliminates token rent but the public sale was preceded by insider allocation at a discount, capital-formation extraction is locked in. If oracle dependence is unaddressed, oracle extraction continues regardless of how clean every other layer is.
The conservation property follows from the seven categories being independent. They exploit different structural asymmetries. Closing one does not close the others. A protocol’s total extractable value is the sum across all seven channels, and reducing one component does not reduce the total unless the others are also reduced.
This has a sharp consequence. Every existing MEV solution achieves zero or near-zero MEV extraction within its scope while leaving the other six channels open. Private relays redistribute MEV but reintroduce platform extraction (the relay becomes the new intermediary). Encrypted mempools defer ordering extraction to the decryption moment but rely on a validator committee that retains governance and platform extraction surface. Solver auctions reduce ordering extraction for matched orders but leave the unmatched orders subject to standard AMM MEV, and introduce solver-network platform extraction. None of these are wrong. They are partial. The conservation property says partial does not compose to total.
Where the argument lands
The constructive corollary is that generalized extractable value must be addressed at every layer simultaneously. There is no architectural shortcut. A protocol that wants extraction to actually be zero must close ordering extraction, governance extraction, token rent, capital-formation asymmetry, oracle extraction, platform extraction, and liquidation extraction, all in the same design.
This is a stronger constraint than the MEV literature typically imposes, and it is the right constraint. The reason existing MEV-resistant protocols are not adopted as substrates for the rest of the DeFi stack is that even within their domain of competence, the other six channels remain open in the surrounding ecosystem. A trader who routes through a perfectly MEV-resistant AMM but borrows on a protocol with governance extraction and uses an oracle with operator extraction has not eliminated their exposure to GEV. They have moved the GEV around.
The architectural pattern that closes the full GEV surface has nine components, which I will describe in subsequent posts of this series. They are not novel individually; commit-reveal batch auctions, uniform clearing prices, Shapley-value distribution, rate-of-change guards, and the other components are established in mechanism design and applied finance. The contribution is compositional: applying all of them simultaneously, at every layer, with composability constraints that prevent extraction from relocating across module boundaries.
Three composability constraints
If GEV-resistance is to survive module composition, three constraints suffice in the architectures I have studied:
First, unified Shapley attribution. Every value flow in the system passes through a single attribution mechanism. If reward distribution is per-module, an attacker can game cross-module boundaries to receive Shapley-violating allocations that no single module’s attribution detects. If attribution is unified, the Shapley axioms apply globally.
Second, unified contribution tracking. Every credit a participant earns is recorded in a single directed acyclic graph. Cross-module contributions are visible to the attribution mechanism, which prevents double-counting in one direction and attribution gaps in the other.
Third, unified safety boundaries. Risk parameters apply globally rather than per-module. A circuit breaker that pauses one module while another remains active creates an exit path through which extraction can flow.
These three are sufficient to preserve GEV-resistance under composition for the protocol architectures I have analyzed. They are not the only sufficient conditions, but they are minimal in the sense that removing any one creates a known extraction path.
What this is not
This is not a claim that MEV is unimportant. MEV is one of the seven channels, and it is the most visible because it operates on millisecond timescales and leaves direct on-chain evidence. The MEV literature has produced real progress on a real channel. The argument is that the literature’s frame is too narrow to be the unit of analysis for protocol design, and that protocols designed to be MEV-resistant alone will discover that the other six channels are still extracting from their users.
This is also not a claim that the seven channels are all that exist. There may be eight or more. The framework’s claim is that the count is greater than one, and that the conservation property holds across whatever the full set turns out to be. Adding channels strengthens the argument; it does not weaken it.
And this is not a claim that any specific allocation rule is universally correct. Shapley value is the unique allocation satisfying efficiency, symmetry, the null-player property, and additivity. Different axiom sets yield different unique allocations. The framework uses Shapley because the four axioms it satisfies are individually defensible on cooperative-game-theoretic grounds. Substituting a different allocation rule, if defensible, would change the specific GEV computation but not the conservation property.
Where the series goes from here
The next post will describe the architectural pattern for closing the seven channels, starting with the order-of-magnitude considerations for why batch auctions with uniform clearing prices are the right substrate for closing transaction-ordering extraction specifically. Subsequent posts will cover the other channels in approximate order of how completely they can be addressed today.
The thesis to keep in mind across the series: extraction is conserved across protocol layers, partial solutions relocate rather than eliminate, and the only architectural pattern that produces zero extraction is one that addresses every channel simultaneously with composability constraints preventing cross-channel relocation.
The argument does not require new mathematics. Shapley value is from 1953. Cooperative game theory has the relevant axiom system. The contribution this series makes is in observing that the abstraction layer for the MEV problem is wrong, and that the right abstraction layer is one that the existing tools can address but the existing protocols have not.
Posted to ethresear.ch under handle schelling. Part 3 of an eight-part series on airgap closure. Comments and counterexamples welcome.