1) Ghost explains: Wide Sandwich Attacks (with updated website at the end)

We looked at a full year of Solana trades and found that wide (multi-slot) sandwiches extracted 529,000 SOL.

Wide sandwiches now account for 93% of all sandwiches 2) A wide sandwich happens when the frontrun and backrun occur in different slots (N -> N+3). This type of sandwich attack evades standard single-slot detectors

Let’s look at Solana MEV for September

-> We’ll also look at Jito bundle compositions

-> A DATASET of 1M atomic arbs is available

Did you know that the top atomic arb searcher on Solana doesn’t use a custom program? They route through Jupiter’s program. More on this in a bit.

Let's start with atomic arbitrage

Tips for atomic arbs hover at around 50% of revenue. This is significantly less competitive compared to Ethereum where you see 90-99%+ paid to the proposer.

Notice also that about half (by revenue) of atomic arbs are sent directly to validators and don’t go through Jito. These are virtually all “blind” or speculative arbs where searchers hope to land a profitable transaction that pays for all the failed ones. Most of these transactions fail but when they hit, they can avoid paying tips or significant priority fees.

Here’s the top arb for September (by CET timezone). The searcher kept 80%. You don’t really see that on Ethereum.

Here’s the 2nd most profitable atomic arb in September and it was sent directly to the validator. It was not part of a Jito bundle. If you check this searcher’s history, you’ll see many failed transactions. To determine profitability you would have to account for the sum of fees for those failed transactions.

Some searchers, like the one above, don’t use a custom program. Instead they route their arbs through Jupiter. We were a bit surprised but also impressed.

See below for average tips per (signer, program) and notice how the profitable searchers use both Jito and direct-to-validator. Many searchers also hardcode their tips. If arb on Solana becomes more competitive, we expect to see more custom programs (to allow for runtime input amount adjustments) as well as dynamic tips


This atomic arb dataset is available to you! Check the end for the dataset and caveats.

Now switching to sandwich MEV:

Notice how tips for sandwiches were significantly lower than for atomic arbs. This kind of MEV is less competitive since it depends on private orderflow from nefarious or misconfigured validators or RPCs through backroom deals.

So where does MEV fit into Jito bundles? Let’s see:

Here, “DEX” means the tip came from a bundle which contained a swap on one of the Solana DEXes, pumpdotfun means the tip came from a bundle with a trade on the bonding curve.

Tips from DEX swaps seem to make up about 50% of total tips, and pumpdotfun makes up about 40%. We split out arb and sandwich tips from the DEX category.

Based on this data, MEV-related tips seem to be lower than tips from users and Telegram bots seeking fast transaction inclusion. However, this doesn't account for CEX-DEX arbitrage tips, which are likely significant. More on that in the future!

We investigated whether bid spoofing was happening on @blur_io and if so, to what extent.

TLDR (part1)

Based on the results of our analysis, @PacmanBlur and the blur team have done a great job disincentivizing bid spoofing

< 0.6% of orders failed due to bid spoofing TLDR (part 2)

> 1m transactions on blur

Of those, we looked at the transactions that attempted to accept collection offers:

~ 465k transactions (1.27m orders)
~ 42k transactions with at least 1 failed order (91k orders failed)

of those, only 8100 orders failed due to spoofing