VaultX research · 2026-11-19
Before VaultX published a single trading signal, we built a real trading bot and tested it the hard way: on a full day of pump.fun data (42.2 million trades, 40,633 new coins), and then live with real money. Most trading bots sold online never show results. These are ours, including the losses.
It listens to the Solana blockchain for new pump.fun coins and new Raydium pools (about 20 new coins a minute), checks each one for rug-pull signs (mint and freeze authority, top-10 holders, RugCheck and GoPlus reports, dangerous token extensions), and buys only when real money flows in fast. Buys go through Jupiter in Jito bundles, and every position has a stop, two take-profits, a trailing stop and a time limit.
We replayed a full day (October 15, 2026) with the bot's own exit rules, pump.fun's fee both ways, tips, and a realistic 2-second buy and 3-second sell delay. Settings were chosen on the first 60% of the day and judged on the last 40%. Every exit grid, single indicator and machine-learning coin score that looked good on the first part lost money on the second.
The best entry ("10+ SOL of real buying within 90 seconds") made money from 00:00 to 08:00 UTC, was flat from 08:00 to 16:00, and lost 9-13% a trade from 16:00 to 24:00, the hours when insiders snipe in the launch block. Outside 00-08 UTC, fast snipes lost 9% to 23% a trade. This is one day of data, so treat it as a warning, not a rule.
After a buy, the median peak was +47%, reached after a median of 31 seconds (a quarter of them within 4 seconds). Holding for a minute or more had a negative median. If you are not out in seconds, the move is usually gone.
Groups of profitable wallets lost 3.5% to 4.8% a trade when copied 2 seconds late, across 6,000-7,700 copies. The single most profitable wallet (+2,717 SOL that day) made +40% a trade, but copying it lost 17% a trade, even half a second late: it wins by being first, so a copier buys its pump and sells into its exit. The top 100 wallets buy in the launch block itself. Only one wallet we tested copied well (+4.6% and +6.8% a trade on two separate parts of the day), and that was picked after seeing the results.
On November 4, 2026 the bot traded live: 311 trades, 82 wins, 229 losses, ending about 1.36 SOL down on 13.9 SOL traded. Most losers never went up after the buy: the delay between seeing a signal and landing a trade is where the money went.
A SOL perpetuals rule that traded only when all 20 indicators agreed on 5-minute candles fired on 4.5% of bars and lost 0.14% a trade before leverage over a year. In our later test across 37 top coins, the same kind of indicator agreement also lost on 1-hour and 4-hour charts, but made a small profit on the daily chart (+0.11R a trade over 887 past setups, about 52% winners).