Summary

Sandwich attacks place a buy transaction before and a sell transaction after a victim’s trade, exploiting slippage to extract value. High-profile data shows this mechanism is widely misunderstood: the top “attackers” by volume often have ~$0 profit or are statistical artifacts. The corrected median price impact from genuine sandwiching is ~0.37 basis points — much lower than the frequently cited 1.53 bps. More MEV value comes from arbitrage and backruns than from direct sandwiching.

Mechanism

[Attacker buy] → [Victim swap] → [Attacker sell]
  1. The attacker observes a pending swap in the mempool.
  2. The attacker buys the same token first, pushing up the price.
  3. The victim’s swap executes at the inflated price (within their slippage tolerance).
  4. The attacker immediately sells, profiting from the temporary price inflation.
  5. Net effect: the victim pays a higher price; the attacker extracts the spread.

Slippage tolerance determines the maximum extractable value per sandwich. A 0.5% slippage tolerance on a $10,000 swap allows the attacker to extract up to $50.

Corrected Statistics

Common claims overstate the frequency and profitability of sandwiching. From the dataalways analysis (Apr 2026):

StatisticCommon claimCorrected
Median price impact1.53 bps0.37 bps
Top “sandwich” by volumeHigh profit~$0 profit
Top 3 “attacks” in the periodSignificant MEV~$0 profit

Why the Discrepancy

  1. Arbitrage misclassified as sandwiching: some automated arbitrage sequences are structurally similar to sandwiches but actually improve price discovery.
  2. Statistical artifacts: bots that appear to sandwich may be executing failed or neutral strategies.
  3. Recency: the most sophisticated sandwiching is done by private systems that don’t appear on public dashboards.

Actual Profitability Distribution

  • The majority of sandbox-labeled transactions have very low or zero net profit.
  • Profitable sandwiches are concentrated among a small number of sophisticated operators.
  • Thin-margin AMM pools (e.g., stablecoin pairs) are unprofitable to sandwich due to low slippage tolerance.

Conditions for Profitable Sandwiching

  1. High slippage tolerance: the victim must be willing to accept a significant price move.
  2. Sufficient liquidity imbalance: the pool must be thin enough that the attacker’s buy meaningfully moves the price.
  3. Gas + latency budget: the attack requires three transactions in precise sequence; high gas costs on L1 make many sandwiches unprofitable.
  4. Public mempool visibility: transactions routed through private RPCs (OFA providers) are not visible and cannot be sandwiched.

Protection Mechanisms

MechanismHow it worksEffectiveness
Private RPCs (OFA, Flashbots Protect)Transaction not visible in public mempoolVery high for covered txs
Low slippage toleranceReduces extractable value; may cause revertsModerate
CoW Swap / batch auctionsTransactions matched in batches at a uniform priceHigh; no in-block ordering advantage
Encrypted mempool (LUCID)Transactions encrypted until after orderingEliminates ordering-based MEV entirely

The $50M UI Lie Case Study

A detailed analysis (Mar 2026) examined a specific case where Aave displayed a “good” swap quote while CoW Protocol’s solver routed to a microscopic Sushi pool with much worse execution:

  • Aave displayed: optimal routing, ~1 bps slippage
  • CoW’s actual execution: routed to a tiny Sushi pool, $50M+ impact on the broader trade flow
  • Root cause: the quote-to-order mismatch — solvers update routing between quote display and actual execution; the UI showed the initial quote, not the final execution

Key insight: the worst execution outcomes are not always from sandwiching. Quote staleness, solver competition failures, and routing to illiquid venues can cause larger damage than a well-executed sandwich.

MEV Taxonomy: Sandwiching vs. Other MEV

By value extracted:

  1. CEX-DEX arbitrage: largest source — price information flows from centralized exchanges to on-chain AMMs
  2. Liquidations: second largest — collateral seized at discount from undercollateralized positions
  3. AMM arbitrage: between on-chain venues
  4. Backrunning: extract residual value after a large trade (without the front-run component)
  5. Sandwiching: smaller than commonly believed; concentrated in specific pools/conditions

Open Questions

  • What is the true rate of “harmful” sandwiching (where the attacker profits at the victim’s expense) vs. statistical sandwiching?
  • Does universal adoption of private RPCs eliminate the sandwich problem, or does it shift value to other MEV types?
  • How do encrypted mempools (LUCID) affect the overall MEV taxonomy — does eliminated sandwich MEV simply shift to other forms?

Key Sources

  • Sandwich Statistics (dataalways, Apr 2026) — corrected 0.37 bps figure; ~$0 profit for top “attackers”
  • The $50M UI Lie (Mar 2026) — code-level analysis of Aave/CoW routing failure; quote staleness