OAK — OnChain Attack Knowledge

Worked example · 2024-06

Solana validator-coordinated sandwich-attack MEV cohort and Foundation delegation-program removal — Solana — 2024-06

Loss
the on-chain MEV-extraction shape on Solana through 2024-2025 is structurally diffuse but at cohort-scale extraction. The single most-publicised cohort instance — DeezNode's private-mempool sandwich-bot operation — extracted ~65,880 SOL (~$13.43M) in a single 30-day window (December 7, 2024 → January 5, 2025), executing 1.55M sandwich transactions with 88.9% success rate. DeezNode alone was responsible for nearly half of all sandwich attacks observed on Solana during the window. Helius's January 2025 MEV report counted sandwich attacks present in 112 of 150 DeFi tokens traded that week, indicating cohort prevalence at the broader-token-universe scale. Aggregate cohort extraction across all Solana validator-coordinated sandwich activity from Jito's public-mempool shutdown (March 2024) through 2025-Q1 is estimated by independent researchers at $50-100M+ in retail-trader MEV losses. The structural extraction shape is operator-side validator-and-private-mempool-cooperation against retail traders, distinct from the EVM mev-boost class because the Solana validator-block-construction layer does not cleanly separate from the private-mempool layer.
Attribution
confirmed. DeezNode is publicly named as the operator of the largest single sandwich-bot program (per Helius MEV report and Solana Compass analysis). Solana Foundation publicly removed 30+ validator operators from its flagship delegation program in June 2024 for repeated sandwich activity; the validator identities are public via the Solana Foundation's delegation-program records and Helius / Solana Compass coverage. The validators-and-private-mempool-operators continue to operate on the network — they retain validator status — but lose the foundation-side stake-subsidy (delegation-program payouts), which is the foundation's primary economic-incentive lever. Cohort-scale attribution to a single super-operator is not established; the structural pattern is consistent with several distinct operator clusters cooperating with the validator cohort rather than a single operator scaling.
Key teaching point
Solana's runtime model does not provide a clean defense against validator-coordinated MEV at the protocol layer; the binding defender-side intervention is foundation-policy. Unlike EVM's proposer-builder separation (which provides a clean trust-substrate boundary that the mev-boost relay layer sits within), Solana's leader-schedule-driven block production gives the active-leader validator full discretion over transaction inclusion and ordering during their slot. The defender-side intervention surface is therefore at the layer above the runtime — Solana Foundation's delegation program, which provides stake-subsidy to validators that meet operational-quality standards. The June 2024 removal of 30+ validators for sandwich activity is the canonical 2024 worked example of foundation-policy as the binding defender-side lever for validator-coordinated MEV. The class is structurally distinct from the EVM T14.002 (mev-boost relay) class where the relay is the trust-boundary; on Solana the trust-boundary is the foundation-policy layer.

OAK Techniques observed:

  • OAK-T9.x (proposed) — Solana-native validator-coordinated MEV extraction, structurally adjacent to but distinct from the EVM OAK-T14.002 (mev-boost-relay-attack) class. The cohort is the Solana-specific analogue of EVM private-mempool / mev-boost MEV extraction but operates within Solana's leader-schedule-driven block-production model rather than EVM's proposer-builder separation. T14.002 is anchored on the EVM mev-boost relay surface; the Solana cohort needs a sibling sub-Technique to capture validator-and-private-mempool cooperation as a distinct primitive.
  • OAK-T5.004 (Sandwich / MEV Extraction) — primary on-chain extraction primitive; each sandwich transaction front-runs and back-runs the victim's trade, extracting the MEV spread between the victim's execution price and the true market price. At cohort scale (1.55M sandwich transactions extracting ~65,880 SOL in a 30-day window), T5.004 is the dominant per-block extraction shape on Solana.
  • OAK-T17.001 (Cross-Venue Arbitrage-Driven Price-Discovery Distortion — the T17 tactic page cross-references sandwich MEV as T17↔T5.004: each sandwich transaction distorts the victim's execution price relative to the true market price, constituting a per-block market-manipulation primitive; at cohort scale (1.55M sandwich transactions extracting ~65,880 SOL in a 30-day window), the aggregate price-discovery distortion across 112 of 150 DeFi tokens is a structural T17.001 surface exercised through validator-and-private-mempool cooperation rather than through conventional cross-venue arbitrage).
  • OAK-T8.001 (Cluster Reuse) at the operator-cluster layer — DeezNode operates a single sandwich-bot program responsible for ~50% of cohort-scale extraction; the program-cluster fingerprint is stable across the 2024-2025 window. Cohort-aggregate methodology (Helius MEV report, Solana Compass scale-or-die analysis) attributes cohort-level extraction to a small set of validator-and-private-mempool operator clusters rather than to a long-tail of distinct operators.
  • OAK-T6 (Defense Evasion) as a layered modifier — the private-mempool surface is structurally opaque to retail traders, who route transactions through wallet / aggregator UX layers that do not surface validator-cluster-level routing risk. The defender-side intervention surface is at the foundation-policy layer (Solana Foundation delegation-program removal) rather than at the protocol-runtime layer.

Summary

Solana's transaction-execution model differs structurally from EVM-side proposer-builder-separation: instead of a public mempool feeding a relay layer that constructs blocks for validators, Solana operates a leader-schedule-driven block-production model where the active-leader validator has direct discretion over transaction inclusion and ordering during their slot (~400ms per slot). The ecosystem's MEV substrate evolved through several phases:

  1. Pre-March 2024 (Jito public-mempool era). Jito Labs operated a public mempool ("Jito Block Engine") that aggregated MEV-bid-augmented transactions and submitted them to leader validators via the Jito-Solana validator client. The public-mempool nature provided some structural transparency to MEV activity but also lowered operator-side coordination costs for sandwich extraction.

  2. March 2024 (Jito public-mempool shutdown). Jito suspended its public mempool, citing concerns about harmful MEV practices including sandwich attacks. The shutdown immediately reduced some MEV-extraction practices but encouraged the rise of alternative private mempools operated by other infrastructure providers, with reduced transparency and exclusive validator-side access.

  3. March 2024 → June 2024 (private-mempool proliferation). Several alternative private-mempool operators emerged, offering validator-side participation in MEV-extraction profits in exchange for transaction-inclusion preference. DeezNode, the largest of these, offered participating validators 50% of the MEV-extraction profits.

  4. June 2024 (Solana Foundation delegation-program removal). The Solana Foundation removed 30+ validator operators from its flagship delegation program for repeated sandwich activity. The removal is a foundation-policy intervention — the affected validators retain their validator status on the network but lose the foundation-side stake-subsidy, which is the foundation's primary economic-incentive lever. The intervention is structurally distinct from a protocol-runtime change.

  5. 2024-Q4 (DeezNode peak window). In the 30-day window from December 7, 2024 to January 5, 2025, DeezNode's sandwich-bot program executed 1.55M sandwich transactions with 88.9% success rate, extracting 65,880 SOL ($13.43M) — approximately $2,200 SOL per day average extraction. DeezNode alone was responsible for nearly 50% of all sandwich attacks observed on Solana during the window.

  6. 2025-Q1 (Helius MEV report). A January 2025 Helius MEV report counted sandwich attacks in 112 of 150 DeFi tokens traded that week, indicating cohort prevalence at the broader-token-universe scale despite the foundation-policy intervention.

The cohort is the canonical 2024-2025 Solana-side worked example of validator-coordinated MEV extraction. The structural pattern — validator-side block-construction discretion + private-mempool-operator-side transaction-inclusion-preference economics + retail-trader-side opaque routing — is the Solana-native analogue of the EVM mev-boost / Flashbots ecosystem but operates within a different protocol-runtime substrate.

Why this is structurally significant

The Solana validator-coordinated sandwich-attack cohort sits at three structurally distinct intersections:

  1. Solana's protocol-runtime model does not provide a clean defense against validator-coordinated MEV. Unlike EVM's proposer-builder separation (which provides a trust-substrate boundary that the mev-boost relay layer sits within and that protocol-side defenses can target), Solana's leader-schedule-driven block production gives the active-leader validator direct discretion over transaction inclusion and ordering during their slot. Protocol-runtime-level interventions (e.g., enforcing block-level ordering invariants) are not currently feasible without substantial runtime changes. The defender-side intervention surface is therefore at the layer above the runtime — foundation policy, not protocol code. This is structurally distinct from EVM's T14.002 (mev-boost relay) class, where the relay layer is the trust-boundary and protocol / governance / vendor changes can target it directly.

  2. Foundation-policy as the binding defender-side lever is a structurally weak-form defense. The Solana Foundation's delegation-program removal is an economic-incentive lever, not a protocol-runtime constraint. Removed validators retain the ability to operate on the network and to participate in private-mempool MEV extraction — they lose only the foundation-side stake-subsidy. The intervention raises the cost of sandwich-MEV participation (the foregone delegation-program payouts) but does not eliminate it. The structural ceiling on this defense is the foundation's economic-incentive footprint relative to the cohort-scale MEV-extraction economics — DeezNode's ~$2,200/SOL/day extraction is high enough that foregoing foundation-side stake-subsidy may be economically rational for the validator-and-operator cohort.

  3. The Solana sandwich-MEV cohort is structurally similar to but distinct from the EVM mev-boost cohort. Both cohorts produce retail-trader MEV losses at industrial scale; both operate through a private-mempool-style routing surface; both have a defender-side cohort (Flashbots Protect on EVM, Jito MEV Protect / Helius / Anza coverage on Solana) that competes for retail flow. The structural distinction is the trust-substrate locus: on EVM, the relay layer is the trust boundary, and changes to relay-side policy can target the cohort directly; on Solana, there is no equivalent relay layer, and the trust boundary is at the foundation-policy layer, which is structurally weaker. The cohort therefore deserves a Solana-native sub-Technique sibling to T14.002, not an extension of T14.002 itself.

Timeline (UTC)

When Event OAK ref
Pre-2024-03 Jito Labs operates public mempool ("Jito Block Engine") with MEV-bid-augmented transaction routing to leader validators; sandwich-MEV operators participate via the public mempool surface (genesis surface)
2024-03 Jito Labs shuts down its public mempool, citing concerns about harmful MEV practices; immediate reduction in some sandwich activity but creates space for alternative private mempools Defender response — vendor-policy
2024-03 → 2024-06 Alternative private mempools emerge; DeezNode, the largest, offers participating validators 50% of MEV-extraction profits in exchange for transaction-inclusion preference (private-mempool proliferation)
2024-06-10 CoinDesk publishes "Solana Heavyweights Wage War Against Private Mempool Operators" coverage of the cohort dynamics and the foundation's emerging response (forensic surface — initial)
2024-06 Solana Foundation removes 30+ validator operators from flagship delegation program for repeated sandwich activity; affected validators retain validator status but lose foundation-side stake-subsidy Defender response — foundation-policy
2024-06 onward Cointelegraph, Coingape, Gate Learn, Bitget News publish secondary-source coverage of the foundation's delegation-program intervention (forensic recurrence)
2024-12-07 → 2025-01-05 DeezNode peak window: 1.55M sandwich transactions, 88.9% success rate, 65,880 SOL ($13.43M) extraction; ~50% of cohort-scale Solana sandwich activity attributable to DeezNode alone T8.001 cohort-scale operator-cluster signature
2025-01 Helius MEV report identifies sandwich attacks in 112 of 150 DeFi tokens traded that week; cohort prevalence at broader-token-universe scale (continuing forensic surface)
2025 onward Cohort persists despite foundation-policy intervention; Solana Compass / Helius / Anza published research continues to track operator-cluster shape; retail-side MEV-protection products (Jito MEV Protect, Helius MEV Protect, others) compete for retail-flow (continuing surface)

What defenders observed

  • Pre-event (operator-cluster layer): the operator-cluster fingerprint is observable on-chain via wallet-cluster analysis on the sandwich-extraction wallets. DeezNode's operator-cluster is publicly identified through its sandwich-bot program signature (a single program responsible for ~50% of cohort-scale extraction). Cohort-aggregate methodology — wallet-cluster analysis + sandwich-pattern correlation across validator slots — is the canonical defender-side analysis pattern; Helius and Solana Compass are the principal vendor-side substrate. Defender lesson: cohort-level analytics are mature; per-incident detection is mature; the binding constraint is the intervention surface, not the detection surface.
  • At-event (block-construction layer): sandwich-MEV is structurally observable on-chain — a sandwich is a triple of transactions (front-run buy, victim swap, back-run sell) within a single validator slot, with the front-run and back-run wallets clustered to a single operator. The signal is detectable with sub-second latency; the cohort-level prevalence statistics (Helius's 112-of-150-tokens figure) are derived from the same signal aggregated across slots. Defender lesson: the at-event signal is mature; the gap is in the user-layer intervention (whether retail traders route through MEV-protected aggregators) and the foundation-policy intervention (whether the foundation maintains the delegation-program removal posture against the cohort).
  • At-event (foundation-policy layer): the foundation's June 2024 delegation-program removal is the canonical defender-side intervention; the affected validators retain validator status but lose foundation-side stake-subsidy. The intervention is observable at the public-record layer (foundation-published delegation-program records) but its operational effect on cohort-scale MEV-extraction is bounded — the cohort persists post-intervention per the December 2024 - January 2025 DeezNode peak window. Defender lesson: foundation-policy interventions raise the cost of cohort-scale MEV participation but do not eliminate it; the structural ceiling is the foundation's economic-incentive footprint relative to the cohort-scale extraction economics.
  • At-event (retail-trader-routing layer): retail traders route through wallet / aggregator UX layers that do not uniformly surface validator-cluster-level routing risk. Phantom, Jupiter, BonkBot, and other major Solana retail-trader-routing surfaces have varying degrees of MEV-protection integration (some through Jito-style protected-transaction submission, others without). Defender lesson: retail-side MEV-protection is a UX-layer surface that requires per-vendor adoption to be cohort-effective; the Solana ecosystem lacks a dominant MEV-protection-default analogous to Flashbots Protect's role on EVM.
  • Post-event (cohort-persistence layer): the cohort persists despite the foundation-policy intervention. The structural reason is the economic-incentive asymmetry: cohort-scale MEV-extraction at ~$2,200/SOL/day is high enough that foregoing foundation-side stake-subsidy may be rational for the operator cluster. Defender lesson: economic-incentive interventions at a single layer are insufficient when the cohort-scale extraction economics dominate the foregone-subsidy economics; the binding intervention requires either (a) a deeper foundation-policy intervention (e.g., direct sanctioning of validator operators), (b) a runtime-level intervention (deeper protocol-runtime change), or (c) a retail-side intervention that diverts cohort-scale flow away from non-MEV-protected routing.

What this example tells contributors writing future Technique pages

  • Solana-native MEV-extraction is structurally distinct from EVM mev-boost MEV-extraction; the OAK Technique surface should reflect this. Future T14.002-adjacent contributors should treat Solana-native validator-coordinated MEV as a sibling sub-Technique, not as an extension of the EVM mev-boost class. The trust-substrate locus — relay-layer on EVM, foundation-policy on Solana — is the load-bearing structural distinction.
  • Foundation-policy is the canonical 2024 Solana-side defender-tooling layer; OAK should record foundation-policy interventions as a first-class observable. Future Solana-native worked examples that involve foundation-policy interventions (delegation-program removal, validator stake-subsidy adjustments, validator client-distribution policy) should preserve the foundation-policy layer as a discrete observable, separate from runtime-protocol changes.
  • Cohort-aggregate methodology generalises across MEV cohorts. The Helius / Solana Compass methodology — wallet-cluster analysis + sandwich-pattern correlation across validator slots — is the v0.x detection primitive for cohort-scale MEV-extraction analysis on Solana. Future cohort-aggregate analysis (Pump.fun rug-republic, M3M3, MEV-extraction) should apply the same methodology.
  • Operator-cluster persistence post-foundation-policy-intervention is informative for defender-tooling priorities. The DeezNode cohort's persistence through the foundation's June 2024 delegation-program removal demonstrates that foundation-policy interventions raise but do not eliminate cohort-scale MEV-extraction costs. Defender-tooling-investment framing should treat foundation-policy as one layer in a multi-layer defense, not as a sufficient stand-alone defense.
  • Distinction from EVM mev-boost cohort matters for the OAK Technique surface. Future Solana validator-coordinated MEV worked examples should explicitly cross-reference EVM mev-boost coverage (see examples/2023-04-mev-boost-equivocation.md and examples/2023-02-jaredfromsubway-mev.md) and preserve the structural distinction at the trust-substrate level.

Public references

Discussion

The Solana validator-coordinated sandwich-MEV cohort is OAK's canonical 2024-2025 worked example for non-EVM-L1 validator-coordinated MEV extraction. The case sits at the intersection of OAK-T14.002 (mev-boost-relay-attack: adjacent EVM Technique), OAK-T8.001 (cluster reuse: at the operator-cluster layer), and OAK-T6 (defense evasion: at the foundation-policy layer).

The structural distinguishing feature of the case is the trust-substrate-locus distinction between Solana and EVM. EVM's proposer-builder separation provides a relay-layer trust-substrate that defender-side intervention can target directly (Flashbots Protect, mev-boost relay-policy changes). Solana's leader-schedule-driven block production does not provide a comparable trust-substrate; the binding defender-side intervention is at the foundation-policy layer (delegation-program removal). This is a structurally weaker defense locus, and the cohort's persistence through the June 2024 intervention provides empirical confirmation.

For OAK's broader cohort coverage, this case + the EVM mev-boost cases (examples/2023-02-jaredfromsubway-mev.md, examples/2023-04-mev-boost-equivocation.md) collectively establish that MEV-extraction is a chain-substrate-class with structurally distinct sub-surfaces per chain ecosystem. The two cohorts converge on the retail-trader-loss-shape but diverge on the trust-substrate-locus and the binding defender-intervention layer. OAK's T14.x v0.x development should preserve this taxonomy and add a Solana-native sibling sub-Technique.

The case also generalises beyond Solana to other non-EVM L1s with leader-schedule-driven block-production models (Sui, Aptos, Sei, Monad). Each of these chains will likely produce its own validator-coordinated MEV cohort as the chain's DeFi ecosystem matures; the Solana case is the canonical 2024-2025 reference for what those cohorts will look like and what the defender-side intervention surface will be. OAK's chain-coverage v0.x work should anticipate this and ensure the framework's MEV-cohort coverage scales beyond the EVM-side framing.

Techniques demonstrated (6)