MEV on Solana: Jito bundles, swQoS, and the anti-sandwich fight

MEV on Solana looks nothing like Ethereum's. There's no native public mempool, block production is leader-based (the assigned leader fully controls their block), and ordering happens through off-chain orderflow and Jito bundles rather than a shared pending-tx pool.

Jito bundles and tip auctions

Jito-Solana (historically ~65% of stake) runs a Block Engine where searchers submit bundles — up to 5 atomic, sequentially-executed transactions, all-or-nothing. An off-chain tip auction (~200ms window) orders bundles by tip and tip-per-compute-unit. A tip is a SOL transfer (min 1,000 lamports) inside the bundle.

searcher → Jito Block Engine:  bundle = [tx1, tx2, ... ≤5], + tip
            ~200ms auction:    rank by tip / tip-per-CU
            leader includes top bundles atomically in its block

Jito also ran a pseudo-mempool that held transactions ~200ms — which leaked orderflow and enabled sandwich attacks (bots front/back-run users into max slippage). After a roughly six-week losing battle, Jito suspended the mempool on 8 March 2024, citing no viable fix. It wasn't re-added — sandwiching simply migrated to private orderflow deals between bots and validators.

swQoS: stake-weighted QoS

Rolled out in 2024, stake-weighted Quality of Servicelets a leader prioritize incoming transaction packets by the stake of the connection: a validator with 1% stake can transmit up to ~1% of packets to the leader. It improves landing odds during congestion and adds Sybil resistance — and it's relevant to MEV because high-stake connections (and the staked-relay services built on them) get privileged transaction-landing lanes.

Paladin: aligning validators against sandwiching

Paladin is a lightweight Jito-Solana fork that rejects sandwich bundles by default and adds P3 (Paladin Priority Port) — a token-gated, rate-limited (~100 TPS) direct-to-leader lane; locking PAL buys a share. PAL has a fixed 1B supply, majority to validators/stakers, designed to make validators financially aligned against sandwiching (a reported ~90% leader / ~5% Paladin stakers / ~5% PAL holders split). It's an economic fix, not a protocol-enforced one.

The honest read

MEV is unavoidable while leaders order transactions — the real question is which MEV. Atomic arbitrage is largely healthy; sandwiching is purely extractive. Every mitigation trades off: killing the mempool didn't end sandwiching (it went private), and Paladin's defense depends on social/economic validator alignment a bot can defect from. There's also a centralization pull — swQoS + private orderflow + co-location all reward high-stake, well-connected validators. Reported sandwich-extraction figures (hundreds of millions over many months) are third-party estimates; treat exact numbers with skepticism.

References

• Jito docs · Paladin docs
• Helius — Solana MEV: an introduction
• Jito block engine, demystified

Solana didn't abolish MEV by removing the mempool — it moved it into the dark. The current fight is over who profits and whether users get sandwiched on the way.