2026-11-03
Trench lessons — 2026-11-03
Key takeaways
- King arbitrage bot (Nov 2): a self-hosted bot's own hot key emptied its wallet 24 s after the bot stopped
- Dashboard PnL can be gross of Jito tips and priority fees, which were 46% of one top pump.fun sniper's gross
- pnl club's 2-of-top-100 signal: wallets that co-buy within 2 s count as one buyer, a 10-min window, and a bigger list didn't help
- FLOAT on Raydium LaunchLab: memecoins quoted in tokenized stocks; 9 of 972 graduated; its own launch had 78 wallets in 20 same-block groups
- Smaller items
King arbitrage bot (Nov 2): a self-hosted bot's own hot key emptied its wallet 24 s after the bot stoppedevidence: A/B: Bitquery chain trace
- What: at 1:32 AM CT Nov 2, a Solana arb bot ("King", running since at least July) lost ~$856K in five plain transfers over 146 s, all signed with its own key, to a never-used address. The bot's last program call was 24 s before the first theft transfer. There was no contract bug. The attacker swapped stablecoins to SOL on Jupiter within 20 min (to dodge issuer freezes), waited ~5 h, and bridged out. Relay and Mayan blocked ~$280K. A second, smaller bot wallet (~$36K) survived because funds were split, and regular sweeps to a withdrawal address limited the loss. Bitquery suggests an alert when the bot stops or money leaves for a new address.
Dashboard PnL can be gross of Jito tips and priority fees, which were 46% of one top pump.fun sniper's grossevidence: C: single X post, one wallet
- What: an on-chain rebuild of a top pump.fun sniper's last 30 days: Axiom showed ~$318K, but after Jito tips ($86K) and priority fees ($62K) the net was $170K. Axiom's number matched PnL *before* those costs "almost to the dollar".
pnl club's 2-of-top-100 signal: wallets that co-buy within 2 s count as one buyer, a 10-min window, and a bigger list didn't helpevidence: C: vendor write-up, no data table
- What: the product tracks the week's top 100 pump.fun wallets and fires when two net-buy the same coin within 10 minutes (≥0.05 SOL each). Wallets that "keep buying the same coins within 2 seconds of each other" are marked a bundle and always count as one buyer. Their claim: single-bundle signals reached 2× "far less often" than two independent buyers, and a 650-wallet list had no higher 2× share than the top 100 (it only produced more signals, ~30–80/day on the free list).
FLOAT on Raydium LaunchLab: memecoins quoted in tokenized stocks; 9 of 972 graduated; its own launch had 78 wallets in 20 same-block groupsevidence: C: X write-ups
- What: FLOAT pairs every coin with an xStock (NVDA/TSLA/SPY…) or a custom quote. Buyers pay SOL and the platform converts. Start mcap is ~$20K, and creators choose a 1/2/2.5% fee. One count shows 972 launches, 9 graduated. A rival's on-chain read of FLOAT's own Oct 28 launch: 133 wallets bought ~56% in ~30 s, including ~78 wallets in 20 same-block groups of near-identical size (~⅓ of supply). The first USDC trader airdrop (35% of fees, by volume) is scheduled for Nov 4 11:00 AM CT, so expect volume-farming wallets on FLOAT coins around then.
Smaller itemsevidence: no action
- GORZ Rug Morgue base rate (B): this week 1,781 coins were graded "Risky" at launch, and 1,315 (74%) died within the hour. 8,221 died with a bundled launch and 6,877 with 6+ same-second sellers. A Risky grade is a strong skip but not a sure thing.
Sources
- https://bitquery.io/investigations/king-arbitrage-bot-hack
- https://x.com/DokuOnChain/status/2117626615341387889
- https://x.com/pnlclublive/status/2117339523948363876
- https://pumpdata.fun/report/2026-11-02.md
- https://pumpdata.fun/stats/2026-11-02.json
- https://x.com/cfm_sol/status/2117489192288805119
- https://x.com/dripstockdotfun/status/2117414167053562155
- https://docs.raydium.io/reference/changelog
- https://developers.jup.ag/changelog
- https://github.com/solana-program/token-2022
- https://solana.com/news/solana-changelog-october-29-2026
- https://solana.com/upgrades
- https://x.com/jujudog69/status/2117362176268669087
- https://x.com/bubblemaps/status/2117633622047940944
- https://gorz.app/morgue
Evidence grades: A = large dataset, B = official docs, C = vendor/thread/small sample. Research notes for education only. Not financial advice.