The ASE 2026-accepted paper examined 200 bot-linked addresses, clustered them by trading behavior and found the biggest profit-rate gap inside a 12-address group routing through Jupiter and proprietary AMMs.
New research on automated trading on Solana found a nearly threefold gap in positive wrapped SOL (WSOL) balance changes inside one small bot cluster. Transactions that invoked HumidiFi, which the paper classifies as a proprietary automated market maker, posted a positive WSOL balance change 62.3% of the time, versus 21.01% for other transactions in the same 12-address group. The comparison covered 244,733 HumidiFi-invoking transactions, including 152,502 with a positive WSOL balance change, and 218,678 other transactions, including 45,949 with a positive change, for a rate ratio of about 2.97. The ASE 2026-accepted paper examined 200 addresses tied to two trading-oriented bot services on Solana, taking the top 100 addresses linked to Trojan and the top 100 linked to SolanaMevBot and collecting their transactions from Oct. 1 to Nov. 1, 2025. It grouped 158 addresses into four clusters using transaction intensity, execution success, fees and asset breadth, while treating 42 as noise. Three clusters, totaling 56 addresses, were treated as MEV-like because transaction-pattern reviews and service labels showed round-trip, cross-venue trading consistent with arbitrage. WSOL appeared in 70.90% to 99.94% of transactions across those groups, but venue mix varied by cluster. The 12-address group at the center of the profit-rate comparison routed heavily through Jupiter and also invoked proprietary AMMs including HumidiFi, with venue IDs mapped from Solscan annotations. The authors said the routes may involve closed or specialized liquidity, but they did not determine why the HumidiFi-linked transactions performed differently. The study’s profit measure was only a pre- and post-transaction WSOL balance change, not a fully netted return after fees, tips, timing, address behavior or route choice, so it shows correlation inside the cluster rather than proof that HumidiFi access caused the gap. A separate 102-address trading-operations cluster concentrated 80.9% of its activity on Pump.fun, underscoring how sharply venue choice can split bot behavior on one blockchain. The paper also analyzed 586 public bot repositories, built separately from the on-chain sample, and its authors published a replication package through Zenodo. CryptoSlate has previously examined Solana’s subsidies for professional trading flow, bot-related congestion and sandwich attacks on retail users, but this study does not connect HumidiFi or the sampled bots to those harms. It also did not compare retail fills, slippage or losses, so it cannot show that public-interface users were directly harmed. The measured transactions nonetheless point to materially different conditions inside the sampled execution paths.