post-migration bounce bot
Building a post-migration bounce bot on Solana (pump.fun) — 2 months, 900+ signals analyzed
Strategy: when a token completes its bonding curve and migrates to pump.fun's pAMM, early buyers dump immediately. Median dump: -72% from migration price. Some tokens bounce. We buy the bottom.
Stack: Python + Helius WebSocket (logsSubscribe by mint, parsing BuyEvent/SellEvent from pAMM program) + GMGN OpenAPI for discovery. We validate every on-chain price using the AMM's constant product invariant (k-gate) to reject secondary pool noise.
What the data shows: time-to-bottom < 25 min is the only feature that consistently predicts recovery (+22pp, n=425). Everything else — smart wallets, holder concentration, wash trading flags — is noise.
Verified on-chain WR after blockchain audit: ~10-15%. We're live at $1/trade fixing data pipeline issues.
The one thing killing us: tokens that bottom in the first 60-90 seconds. We subscribe to Helius after migration is confirmed, so the entire dump happens before our subscription is active. By the time we're watching, the bounce is already 150%+ done and we're buying the top.
Has anyone solved early subscription? Specifically: do you subscribe to AMM events during the bonding curve phase (before migration), and if so — how do you decide which of the hundreds of daily tokens are worth the subscription overhead?