Agents
Solana rug pulls are predictable from the first five minutes of trading across 6.4M tokens
An August 20 arXiv paper tracked 6.4 million Solana tokens over seven months and found most memecoins exhibit rug pull characteristics within one hour of launch, making early detection the only useful kind. Gradient boosting (XGBoost) on the first five minutes of trading data delivered strong prediction without needing any code-level features, and fusing data across PumpFun and Raydium substantially mitigated domain shift between the two venues. Prior rug pull work concentrated on Ethereum, so the Solana liquidity-manipulation pattern was largely unmodeled.
Source
↳ Follow the thread