Skip to content

Priority fees and MEV

How Solana priority fees work in Candle, and what Candle has and has not measured about sandwiching. Not investment advice.


On Solana, a transaction can bid a priority fee, in lamports, to be included sooner (Solana fees). Candle’s agent API asks the venue for its veryHigh estimate on jupiter-venue builds, up to a ceiling (Agent trading):

SOL leg of the tradeDefault ceiling
up to 0.04 SOL100,000 lamports (the floor)
0.04 to 4 SOL25 bps of the leg
4 SOL and above10,000,000 lamports
neither leg is SOL10,000,000 lamports

maxPriorityFeeLamports (0 to 10,000,000) replaces the ceiling for one build. When the ceiling binds, the transaction bids less than the venue wanted and may expire before inclusion. An expired transaction moves nothing.

A sandwich is a trade placed just before yours (pushing the price up) and another just after (selling into it), costing you the difference up to your slippage tolerance. The tolerance is the attacker’s budget.

What Candle sends. Solana swaps are built through Jupiter (raced against DFlow) and sent to a standard RPC. Candle does not use Jito bundles or a private relay for swaps. On Hood, transactions go to a single first-come-first-served sequencer with no public mempool, which makes a front-run structurally impractical (inferred from vendor documentation).

What Candle measured (2026-10-25).

  • 0 of 60 Solana agent trades scanned block by block had a same-signer front-run and back-run on the same token in the same slot.
  • 0 of the 8 worst-filled trades showed a wider sandwich across the slots before and after.
  • Across the five matched Hood trades, zero same-trader sandwiches were detected within two blocks: no trader appeared both before and after a matched trade on the same token. Adjacent trades did occur, including a $55 sell before one trade and $57 and $481 trades after another, by different traders.
  • A control scan found the pattern does exist on Solana in general, so the detector can find it.

Limits. Detection is heuristic, the sample is small (60 trades, 8 for the wide check), and an attacker using two unrelated wallets would be missed. “No sandwich found” is not “no sandwich possible”.

The exposure is loose slippage. If every trade were sandwiched to its tolerance, the most that could be taken from 30 days of Solana agent volume (404 SOL) is about 18.8 SOL. Measured buy-side shortfall against the quote from all causes was 0.72 SOL. Keep tolerance tight; see Slippage.