OAK — OnChain Attack Knowledge

Worked example · 2023-20

Multi-block MEV TWAP oracle grinding via proposer-builder coordination — Ethereum L1 — 2023–2025

Loss
aggregate mid-seven-figures USD across the multi-block MEV TWAP-manipulation cohort spanning 2023–2025. The extraction is distributed across multiple lending protocols and DEX TWAP-consuming contracts where the manipulated TWAP oracle price deviated from the true market price sufficiently to trigger adverse liquidations, unfavourable swap executions, or collateral-ratio manipulations. Individual incident extractions range from low-five-figures (small-scale TWAP deviation exploited against a single isolated lending position) to mid-six-figures (multi-block sequences where the proposer-controlled price grinding produced a cumulative TWAP deviation large enough to exploit a material lending position). The aggregate extraction is a conservative reconstruction from MEV analytics platforms (EigenPhi, MEV-Explore), per-protocol TWAP-deviation analysis, and Flashbots Research multi-block MEV characterisation.
OAK Techniques observed
OAK-T17.005 (TWAP Oracle Manipulation via Multi-Block MEV — primary; the load-bearing manipulation primitive is proposer-controlled multi-block sequencing to grind a TWAP oracle toward a target price across consecutive proposer-controlled slots). OAK-T9.001 (Oracle Price Manipulation — co-occurring; the upstream price-manipulation primitive, staged across multiple proposer slots rather than within a single transaction). OAK-T5.004 (Sandwich / MEV Extraction — structurally adjacent; the multi-block MEV searcher's per-block execution involves sandwich-like trade placement, but the multi-block staging across proposer-controlled blocks is the T17.005-distinctive surface).
Attribution
pseudonymous The multi-block proposer-control sequences are attributable to specific builder entities through MEV-Boost relay bid-traces and beacon chain proposer-assignment data (fee-recipient addresses). The dominant builders (beaverbuild, Titan, rsync, Flashbots builder) are publicly identified; the specific searcher addresses behind individual multi-block manipulation sequences are pseudonymous. Attribution is at the builder-entity-concentration layer rather than at the named-individual-searcher layer.
Key teaching point
Post-Merge Ethereum builder concentration — where a small number of builders (beaverbuild, Titan, rsync) propose a material share of consecutive blocks — is the structural pre-condition for T17.005 multi-block TWAP grinding. A single builder that already proposes a material fraction of blocks can secure consecutive-block proposer sequences without additional coordination cost, making multi-block manipulation an architectural feature of the proposer-builder-separation design rather than an attacker-orchestrated condition. The T17.005 surface is not gated by the attacker's ability to bribe or coordinate proposers — it is gated by the baseline builder-concentration level, which is a structural property of the MEV-Boost builder market.

Summary

Post-Merge Ethereum (September 2022 onward) introduced proposer-builder separation (PBS) via the MEV-Boost relay network. Under PBS, validators outsource block construction to specialised builders who compete in an auction to have their blocks included. The builder market quickly concentrated: a small number of builders — beaverbuild, Titan, rsync, Flashbots builder — won a material share of blocks, particularly during periods of high MEV activity. Builder concentration means that a single builder already proposes a material fraction of consecutive blocks, making multi-block proposer-control sequences a structural feature of the MEV-Boost ecosystem rather than an attacker-orchestrated condition.

Flashbots Research characterised the multi-block MEV attack surface in 2021–2022 publications [flashbotsmultiblock2022], establishing that proposer control over consecutive blocks enables manipulation strategies not feasible within a single block. TWAP oracle manipulation was identified as a canonical multi-block MEV attack class: the attacker spreads price movement across multiple consecutive blocks with smaller per-block capital requirements, achieving a cumulative manipulation effect that exceeds the single-block-constrained manipulation amplitude.

The 2023–2025 observation period documented operational multi-block TWAP grinding on Ethereum mainnet. MEV analytics platforms (EigenPhi, MEV-Explore) and builder-concentration dashboards (mevboost.pics, relayscan.io) surfaced sequences where a single builder proposed 3–5 consecutive blocks, and during those blocks, the TWAP-input venue (a Uniswap V3 pool, typically) experienced correlated price movement in a consistent direction. The per-block price changes were individually modest (1–3% per block) — below typical single-block manipulation detection thresholds — but the cumulative TWAP deviation across the proposer-controlled block sequence was large enough (5–10% deviation) to cross liquidation thresholds on lending protocols that consumed the TWAP oracle.

The extraction pattern composes: builder concentration (structural PBS property) → multi-block proposer control → per-block TWAP-grinding trades → cumulative TWAP deviation → lending-market oracle consumption → liquidation or adverse-execution extraction. The attacker's per-block capital requirement is lower than single-block manipulation because the manipulation is spread across multiple blocks; the constraint is proposer-control (how many consecutive blocks the builder proposes) rather than capital (how much can be borrowed in a flash loan).

The Inverse Finance TWAP manipulation (April 2022, ~$15.6M) is a dual-classified anchor (T9.001 + T17.004 + T17.005 adjacent) where the multi-block execution pattern is structurally adjacent to T17.005 — the manipulation was executed across blocks rather than within a single atomic transaction [examples/2022-04-inverse-finance-twap.md]. Post-Merge (2023–2025), the T17.005 surface has been characterised at the builder-concentration and protocol-oracle-consumption layers, though named-attacker per-incident anchors at extraction-scale comparable to Inverse Finance have been sparse in the public record — a structural property of the builder-concentration attribution layer where extraction is attributed to builder entities rather than to named individual searchers.

Timeline (UTC)

When Event OAK ref
2021–2022 Flashbots Research characterises multi-block MEV attack surface; TWAP oracle manipulation via multi-block proposer control identified as canonical multi-block MEV class (T17.005 class characterised in research literature)
2022-09 Ethereum Merge; proposer-builder separation via MEV-Boost relay network introduced; builder market begins concentrating (T17.005 structural pre-condition created — builder concentration)
2022–2023 Builder concentration data emerges: beaverbuild, Titan, rsync, Flashbots builder win material share of blocks; mevboost.pics and relayscan.io surface per-builder block-proposal share dashboards (builder-concentration surface documented)
2023–2024 Operational multi-block TWAP grinding observed on Ethereum mainnet; EigenPhi and MEV-Explore surface proposer-concentration + correlated per-block price movement sequences; cumulative TWAP deviations cross lending-market thresholds T17.005 (multi-block TWAP grinding via builder concentration)
2023–2025 MEV analytics community develops proposer-concentration + TWAP-deviation joint monitoring methodology; per-protocol oracle-consumption analysis identifies TWAP-vulnerable lending markets (defender-side methodology emergence)
2025 onward Builder concentration persists as structural PBS property; multi-block TWAP grinding remains a standing T17.005 surface; the mitigation surface is at the TWAP-window-length / multi-venue-oracle / per-protocol-deviation-alerting layer T17.005 (ongoing surface)
Continuing Flashbots Research, academic literature, and audit-firm advisories continue to characterise multi-block MEV as a structural PBS property; named-attacker per-incident anchors at extraction-scale remain sparse — a structural property of the builder-concentration attribution layer (research + documentation surface)

Realised extraction

Aggregate mid-seven-figures USD across the cohort (2023–2025). Per-incident extraction is bounded by (a) the number of consecutive proposer-controlled blocks (typically 3–5 for builder-concentration-driven multi-block sequences), (b) the per-block price-movement amplitude achievable without triggering single-block manipulation detection, (c) the lending market's TWAP-consumption window (shorter TWAP windows are more vulnerable to multi-block grinding), and (d) the size of the exploitable position (typically a lending-market position whose liquidation or collateral threshold crosses at the manipulated TWAP). No single extraction-event at Inverse Finance scale (>$15M) for a pure-T17.005 named incident has been publicly documented post-Merge; the aggregate extraction is distributed across smaller multi-block sequences.

Public references

  • Flashbots Research — multi-block MEV characterisation (2021–2022); TWAP oracle manipulation as canonical multi-block MEV class — [flashbotsmultiblock2022]
  • EigenPhi — MEV analytics data on proposer-concentration and multi-block patterns (2023–2025) — [eigenphi2023mev]
  • MEV-Explore — MEV transaction data and builder-attribution dashboards
  • mevboost.pics — per-builder block-proposal share dashboards; builder-concentration data (2022 onward)
  • relayscan.io — MEV-Boost relay data; proposer-assignment and builder-bid traces
  • beaconcha.in — beacon chain proposer-assignment data; fee-recipient attribution
  • Cross-reference: T17.005 at techniques/T17.005-multi-block-mev-twap-oracle-manipulation.md
  • Cross-reference: examples/2021-2025-multi-block-mev-twap-cohort.md — Multi-block MEV TWAP oracle manipulation cohort (2021–2025)
  • Cross-reference: examples/2024-2025-jupiter-dca-solana-twap-oracle-manipulation.md — Jupiter DCA TWAP oracle manipulation on Solana (2024–2025)
  • Cross-reference: examples/2022-04-inverse-finance-twap.md — Inverse Finance TWAP manipulation (~$15.6M; T17.004 + T17.005 adjacent)
  • [daian2020flashboys] — Daian et al. 2020, "Flash Boys 2.0" — foundational MEV characterisation
  • [zhou2023sok] — Zhou et al. 2023, academic taxonomy including TWAP-manipulation classification
  • [ethereumpbs2022] — Ethereum proposer-builder separation specification; structural builder-concentration surface

Techniques demonstrated (3)