New analysis into automated buying and selling on Solana discovered a virtually threefold distinction in constructive wrapped SOL outcomes inside one small group of bots. Transactions that invoked HumidiFi, which the paper classifies as a proprietary automated market maker, recorded a constructive WSOL stability change about 62.3% of the time. Different transactions in the identical 12-address group did so at a 21.01% fee.
The comparability coated 244,733 HumidiFi-invoking transactions, together with 152,502 with a constructive WSOL stability change, and 218,678 different transactions, together with 45,949 with a constructive change. The speed ratio was about 2.97.
The sting appeared in a single slender cluster
The ASE 2026-accepted paper examined 200 addresses related to two trading-oriented bot companies on the Solana community. The researchers chosen the highest 100 addresses linked to Trojan and the highest 100 linked to SolanaMevBot, then collected their transactions from Oct. 1 to Nov. 1, 2025. They grouped 158 addresses into 4 clusters utilizing measures reminiscent of transaction depth, execution success, charges, and asset breadth, whereas treating 42 addresses as noise.
The paper additionally studied 586 public bot repositories, however that software program pattern was constructed individually and can’t be linked to the 200 on-chain addresses. Its authors revealed a replication bundle by way of Zenodo.
Transaction-pattern critiques and repair labels supported treating three clusters, totaling 56 addresses, as MEV-like. On this context, that meant the bots confirmed round-trip, cross-venue buying and selling patterns per arbitrage. WSOL appeared in 70.90% to 99.94% of transactions throughout these teams, however every cluster used a distinct venue combine.
The 12-address group on the heart of the profit-rate comparability routed closely by way of Jupiter and in addition invoked proprietary AMMs together with HumidiFi. The researchers recognized venues from the applications referred to as inside every transaction and mapped these program IDs utilizing Solscan annotations. They mentioned the routes might contain closed or specialised liquidity, however didn’t decide why the HumidiFi-linked transactions carried out in a different way.
The research’s revenue measure was a change between pre-transaction and post-transaction WSOL token balances. It was not a totally netted account of technique returns that remoted charges, suggestions, timing, tackle conduct, route selection, or different potential variations. The information present a correlation inside this cluster, not proof that HumidiFi entry brought about the hole.
A separate 102-address trading-operations cluster concentrated 80.9% of its exercise on the Pump.enjoyable ecosystem. Venue selection sharply separated the sampled bots’ exercise, even on one blockchain.
CryptoSlate has beforehand examined Solana’s subsidies for skilled buying and selling move, bot-related congestion, and sandwich assaults on retail customers. This research doesn’t join HumidiFi or the sampled bots to these harms.
For atypical DEX customers, the discovering means that contributors on one chain can encounter totally different alternatives when their routing, infrastructure, or venue set differs. The research didn’t evaluate retail fills, slippage, or losses, so it can’t present that public-interface customers had been immediately harmed. The measured transactions nonetheless reveal materially totally different situations contained in the sampled execution paths.
