pump.fun On-Chain Trading Dataset for Pump-and-Dump Detection (April–May 2026)

This dataset contains on-chain trading data collected from the pump.fun bonding-curve token launchpad on the Solana blockchain, covering the period 28 April 2026 to 29 May 2026. It was collected to support a supervised machine learning study on early detection of pump-and-dump (P&D) schemes. The dataset comprisesA SQL database of pump.fun tokens with three components: (1) token-level metadata for 110,469 tokens observed during the collection period; (2) 20,731,091 individual on-chain trade records; and (3) pre-computed checkpoint feature vectors for 60,890 in-scope tokens across seven vSOL checkpoint levels (40, 45, 50, 55, 60, 65, 70 vSOL), including binary P&D labels assigned by an automated labeller calibrated through human annotation. Data was collected via the Helius RPC WebSocket and stored in a SQLite database. Graduated tokens (vsol_peak ≥ 115 vSOL) and Mayhem Mode tokens are excluded from the supervised modelling set.