Summary
LlamaRisk supports expanding Chainlink SVR system to Aave v3 deployments on Base and Arbitrum. The proposal activates SVR feeds for 12 assets on Base and 12 assets on Arbitrum in a single phase, enabled by Chainlink’s adaptation of the Atlas framework for L2 networks.
The Ethereum deployment has produced clear and measurable results. In roughly nine months of operation, SVR has processed approximately $675 million in liquidation volume across around 3,900 events, recapturing approximately $16 million in oracle extractable value (OEV). The average recapture rate has reached 73%. The system has operated without incident through significant stress events, most notably the October 10, 2025 market crash in which more than $250 million was liquidated in a single day.
The architectural adaptation for L2s replaces the Flashbots MEV-Share component with Atlas, a smart contract framework that settles auctions onchain rather than solely relying on block builder simulations. This change is necessary because the block building dynamics on Base and Arbitrum do not support reliable offchain simulation natively. It is notable that simulations are still performed offchain as a preliminary phase, where the bids are ordered in terms of size and sanity checks are executed. The core trust assumptions of SVR are preserved: price updates continue to originate from Chainlink DON nodes, and the fallback mechanism mirrors the Ethereum configuration.
1. Current State of SVR on Aave Ethereum
1.1 Financial Performance
SVR has been live on Aave Ethereum since March 2025 and has expanded through three phases to cover the majority of the protocol’s liquidation-relevant collateral. Based on data from our SVR monitoring dashboard at svr.llamarisk.com, the system has produced the following aggregate outcomes through early March 2026:
- Cumulative liquidation volume processed via SVR: approximately $675 million
- Total OEV recaptured: approximately $16 million
- Number of SVR liquidation events: approximately 3,900
- Average recapture rate: 53% of maximum available OEV
Revenue from recaptured OEV is distributed as follows: 65% to the Aave ecosystem, 31.5% to Chainlink, and the remaining 3.5% to block builders. In absolute terms, Aave has received approximately $10.4 million and Chainlink approximately $5.0 million from SVR since inception.
Source: LlamaRisk SVR Dashboard, March 14, 2026
The recapture rate has improved substantially as the searcher pool has grown and competition has increased. Early auctions after each phase activation began at lower recapture rates before converging upward as searchers adapted. As of now, 93 unique searchers have won the Flashbots auctions and completed the SVR liquidation actions, indicating a healty and competitive recapture environment.
Source: LlamaRisk SVR Dashboard, March 14, 2026
1.2 Stress Event Performance
The most significant test of the system to date was the market crash on October 10, 2025, which produced more than $250 million in single-day liquidations on Aave across all markets. SVR continued to function normally throughout the event. No bad debt was generated, no fallback failures were observed, and the system handled the elevated throughput without degradation in recapture rates. A subsequent stress period in late January to early February 2026 produced another round of heavy liquidation volume, again without systemic issues.
These events are meaningful evidence that the system’s design is robust under the exact conditions that matter most: large, rapid market moves where oracle update frequency increases, gas costs spike, and liquidation opportunities concentrate in short windows.
1.3 Coverage Scope
SVR feeds currently cover approximately 75% of Aave’s total Ethereum TVL. The asset list spans both Core and Prime Ethereum markets and includes major collateral types across the BTC, ETH, and governance token categories.
Source: LlamaRisk SVR Dashboard, March 14, 2026
Recapture rates vary by asset. Assets with higher liquidation bonuses and deeper on-chain liquidity tend to attract more competitive bidding, resulting in higher effective recapture rates. Assets with thinner liquidity or smaller liquidation bonuses show lower average rates, partly due to fixed gas overheads consuming a larger share of the available OEV.
1.4 Technical Reliability
The fallback mechanism, which reverts the SVR feed to the standard Chainlink Price Feed after a configurable cutoff time when the SVR route is unavailable, has not been observed to trigger in ways that indicate infrastructure problems. The current cutoff on Ethereum is 6 blocks. This configuration has provided a clean separation between the auction window and the availability guarantee, with no material delays in price delivery to the protocol.
Source: LlamaRisk SVR Dashboard, March 14, 2026
2. Atlas Architecture for Base and Arbitrum
2.1 Why a Different Architecture Is Required
SVR on Ethereum relies on Flashbots MEV-Share to coordinate the liquidation auction. Searchers monitor a private mempool for SVR price updates, submit bundles pairing a liquidation with the oracle update, and block builders simulate each bundle to select the winner. This simulation-based selection is accurate on Ethereum because block builders have deterministic access to current state and can reliably pick the highest valid bid before any transaction reaches the chain.
This approach does not translate to L2 environments. On Base and Arbitrum, block production is controlled by a sequencer rather than a competitive builder market. The simulation environment available to offchain actors is materially less reliable than on Ethereum mainnet, which means off-chain winner selection cannot be trusted to the same degree. MEV-Share itself does not operate on these networks, so there is no equivalent private bundle infrastructure to build on.
2.2 How Atlas Works
Chainlink’s solution is to move auction settlement onchain using Atlas, a smart contract framework originally developed by Fastlane and since incorporated into Chainlink’s infrastructure. The core idea is straightforward: rather than determining the auction winner through offchain simulation, the winner is determined by executing bidders sequentially inside a try/catch loop onchain until one succeeds.
The flow of an Atlas SVR auction on Base or Arbitrum proceeds as follows. When a Chainlink DON node determines that a price update must go onchain (triggered by either a deviation threshold or heartbeat), it delivers the price report to Fastlane’s Operations Relay, which submits the auction. Solvers (liquidators) monitor the relay for new opportunities, analyze the available OEV, and submit signed solver operations with associated bids. The relay collects these bids, orders them by bid amount, and selects the top bids for inclusion. A Chainlink DON node acting as the bundler then assembles a single Atlas transaction containing both the oracle update and the ranked solver operations, and submits that transaction directly to the sequencer or public mempool.
The Atlas contract executes the solver operations in bid order. Each solver gets a chance to execute its liquidation; if it fails to pay the stated bid within its allocated gas, its operations are reverted and the next solver is tried. The first solver to succeed pays its bid and the transaction completes. Because the Chainlink node itself submits the transaction, the sequencing of the oracle update and the liquidation cannot be disrupted by a third-party RPC or sequencer reordering.
An important property of this design is that the atomicity of the Atlas transaction, combined with the DON node sending it directly, provides an ordering guarantee that does not depend on any private mempool infrastructure. The oracle update and the winning liquidation are either included together or not at all.
2.3 Gas Accounting and Solver Requirements
Atlas introduces a specific accounting model for gas costs. Solvers must bond native token (ETH on Arbitrum, ETH on Base) in advance to cover the gas consumed by their solver operations. This bonded balance, denominated in atlETH, is used to repay the Chainlink node that fronted the gas cost for the full transaction.
If a solver’s operation is executed and succeeds, it pays the gas cost of its own operation plus a share of the overall transaction gas. If its operation is executed and fails, it pays only the gas cost of its own failed operation. If it is not executed at all, it pays nothing. This structure ensures that failed competitors do not impose undue gas costs on the winning solver, and that the Chainlink node is made whole regardless of outcome. The only requirement for auction participation is bonding enough native token to cover one transaction’s worth of gas; there is no capital requirement tied to bid size or protocol behavior.
2.4 Fallback and Continuity
The fallback mechanism for Atlas-based SVR is identical in design to the Ethereum implementation. A time-based parameter (s_cutoffTime) determines how long the system waits for an SVR auction to settle before falling back to the standard on-chain price feed update through the public mempool or sequencer queue. BGD Labs has proposed keeping this parameter at 60 seconds for the initial activation on Base and Arbitrum, matching the Ethereum configuration, even though the technical minimum could be lower on L2s. This is a conservative choice that reduces the sensitivity of the system but could be improved significantly as the system scales.
The requireFulfillment parameter is set to true, meaning that if all solver operations fail and no liquidation takes place after an oracle update, the price report reverts rather than being delivered without a corresponding liquidation capture. In such a case, the price update is pushed through a fallback feed as the fallback window expires.
2.5 Immutability and Audit Status
The Atlas contracts are deployed as immutable, publicly accessible contracts and have undergone multiple independent audits (Certora, Cantina, Spearbit). The onchain settlement model has the additional property that its behavior is fully inspectable: any observer can reconstruct exactly which bids were submitted, in what order they were tried, and which one succeeded. This is a meaningful transparency improvement relative to the offchain simulation-based model used on Ethereum, where the selection process happens inside private builder infrastructure.
2.6 SVROracleSteward
As with all previous SVR activations on Ethereum, this expansion includes the deployment of SVROracleStewards for each new feed. The steward contract gives the Aave Protocol Guardian the ability to revert any individual SVR feed back to the standard (non-SVR) Chainlink price feed at any time without a governance vote. This provides a direct circuit breaker in the event of a critical failure or unexpected behavior in any specific feed. Its scope is deliberately narrow: it can only replace an SVR feed with the feed that was already in production, not with any arbitrary oracle, so it does not introduce new oracle risk.
Section 3: Observed Behavior and Risk Considerations
3.1 Liquidator Selection Pressure: Preference for High-Bonus Collateral
One pattern that emerges clearly from observed SVR auction data is that liquidators concentrate their liquidations on collateral with the highest liquidation bonuses in a liquidatable mixed collateral position, if a close factor of 100% has not kicked in yet. Otherwise, when a full liquidation becomes possible, the liquidation calls will still be sorted in the preference of profit. This is rational behavior: a higher liquidation bonus represents a larger gross spread available to the liquidator, which in turn supports higher absolute bids while leaving the liquidator profitable. In competitive auctions, a larger spread means that the competition for those positions is more intense.
The practical consequence is that recapture rates for high-bonus collateral types tend to be structurally higher than for low-bonus assets. Collateral with bonuses in the 5-10% range attracts aggressive bidding, with recapture rates often approaching or exceeding 80%. Assets with bonuses in the 2-4% range see more modest competition, and recapture rates reflect the thinner margins available.
This concentration effect has implications for how the asset scope of SVR activations should be evaluated. The incremental OEV benefit of adding a new asset to SVR is not uniform: assets with high liquidation bonuses and significant collateral usage in Aave contribute disproportionately to total recaptured value. Similarly to Ethereum Core, on both Base and Arbitrum, the proposed asset lists include a mix of high-bonus assets such as AAVE and wrapped BTC variants, and lower-bonus stable collateral.
3.2 Layered Backruns: SVR Backruns Being Backrun
A more subtle phenomenon observed in the Ethereum data is that SVR liquidation transactions can themselves be backrun by other searchers in the same block. This occurs when a searcher only focuses on a subset of all possible liquidations enabled by the SVR price feed update, especially during the high volatility periods
What makes this observation noteworthy from the perspective of SVR design is that these secondary backruns, while not initially reasoned to be captured by the SVR mechanism, still represent a layered OEV on the output of the SVR feed update. More importantly, they do not extract value from Aave or its users; the OEV from the original liquidation has already been recaptured by the SVR auction before the secondary backrun occurs.
In cases where the secondary backrun is itself a liquidation on an SVR-enabled feed, the value flows back into the SVR recapture mechanism. This creates a layered structure where an initial oracle update triggers primary liquidation(s) via SVR, after which there remains OEV to be captured via subsequent liquidations. Both layers generate recaptured OEV. The existence of these layered recaptures is a sign of a mature and efficient searcher ecosystem: it indicates that searchers are monitoring execution output closely, such that no OEV is left uncaptured when the oracle price state changes.
Source:
Etherscan, March 14, 2026
3.3 Risk Considerations
Searcher Pool Depth on Base and Arbitrum
The most substantive risk in this expansion is the early-stage depth of the SVR searcher pool on Base and Arbitrum. New networks start with fewer searchers, less calibrated bidding logic, and lower competition, which translates directly into lower recapture rates during the ramp-up period.
This is not a reason to delay the expansion, but it is a factor that should be tracked closely. A lower SVR participation or a recapture rate in the early weeks after activation does not indicate a system failure, it indicates a network effect that requires time to develop. The fallback mechanism ensures that liquidations are never blocked regardless of auction outcome, so protocol safety is not a function of searcher pool depth.
The practical risk is closer to an opportunity cost: OEV that could theoretically be recaptured is instead retained by liquidators during the early period. This is structurally similar to the trajectory observed on Ethereum after each of the three activation phases, where recapture rates started low and increased as searchers were onboarded and added support for the newly covered feeds. For Base and Arbitrum, the Atlas-specific SDK and integration requirements may extend this ramp-up period relative to Ethereum, since searchers need to implement different tooling for the on-chain settlement mechanism.
On-Chain Settlement Gas Overhead
Atlas introduces a fixed on-chain settlement cost of approximately 250,000 gas per auction transaction, regardless of how many solver operations are executed. For smaller liquidation events where the OEV is modest, this overhead can represent a material fraction of the available value and may reduce net recapture below what would be possible with off-chain settlement. For large liquidation events, the fixed overhead is negligible relative to the OEV at stake.
On Base and Arbitrum, gas costs are substantially lower than on Ethereum, which partially offsets this concern. The per-auction overhead in dollar terms will be a small fraction of what the same gas cost would represent on mainnet. The more relevant consideration is the marginal cost of executing multiple failed solver operations in a competitive auction with many bidders. The gas accounting model in Atlas charges failed solvers for their own operations, but the winning solver still bears the cost of the full successful execution. In practice this is managed through appropriate bid sizing, but auction design efficiency should be monitored as the searcher pool grows and competition increases.
Conclusion
SVR has demonstrated, through roughly a year of live operation on Ethereum, that it can function reliably as a liquidation recapture mechanism across both normal market conditions and stress events. The cumulative $16 million in recaptured OEV, delivered without protocol incidents or bad debt, provides a solid foundation for expanding to additional networks.
The Atlas-based architecture for Base and Arbitrum is a reasonable engineering response to the constraints of L2 environments. It preserves the same trust assumptions as the Ethereum system while eliminating the dependency on private block builder infrastructure that is not available on these chains. The on-chain settlement model is audited, immutable, and more transparent in its execution than the off-chain simulation approach used on Ethereum.
The main variable to track after activation is searcher participation on the new networks. Recapture rates will start below the Ethereum averages and converge upward as competition develops. The fallback mechanism provides a safety floor for protocol operations independent of auction efficiency, and the SVROracleSteward provides a rapid response capability in the event of a feed-level issue.
Disclaimer
This review was independently prepared by LlamaRisk, a community-led decentralized organization funded in part by the Aave DAO. LlamaRisk is not directly affiliated with the protocol(s) reviewed in this assessment and did not receive any compensation from the protocol(s) or their affiliated entities for this work.
The information provided should not be construed as legal, financial, tax, or professional advice.
