2026-10-30
Trench lessons — 2026-10-30
Key takeaways
- Binance hack (Oct 29): withdrawals frozen including Solana routes, SOL under $100, pump.fun activity well below average
- Raydium CPMM InitializeV2 creator fees are not live on mainnet yet: check a program's deploy slot before relying on a changelog
- pumpdata.fun: a free daily pump.fun stats file with wallet classes, and its top winners are not a copy list
- CATBOT post-mortem: curve skipped, tokens handed out, 99.95% off the top
- Bloom/TradeWiz follow-up: drained addresses get drained again
Binance hack (Oct 29): withdrawals frozen including Solana, SOL under $100, pump.fun activity downevidence: B: Binance notice; C: news
- What happened: Binance says about $6.3B left its hot wallets, roughly $3B is frozen on-chain, and losses will be covered from its SAFU fund. Per Binance, a compromised engineer laptop gave access to a service that could submit withdrawals. Withdrawals were halted at 22:15 UTC Oct 29, with the Solana-network SOL/USDC/USDT routes among the first paused. Binance expects withdrawals to resume within 48 hours of 04:17 UTC Oct 30. Deposits still work.
- Market: no SOL was stolen, but SOL fell from $98.94 to $95.64 in under an hour, and about $628M of positions were liquidated in one hour (CoinGlass via CryptoTicker). pump.fun on Oct 29 (pumpdata.fun recorder): 752 graduations, 34% under the 30-day average; SOL traded 22% under; launches 13% under. That covers the full UTC day, so it's a slow day plus the evening shock.
- Lesson: treat the freeze window as abnormal when you score wallets or strategies. Don't write off a wallet as dead from those hours alone, and expect a burst of brand-new exchange-funded wallets once withdrawals reopen, which can look like fresh or linked wallets to funding checks. Loss limits set in SOL don't move when SOL's dollar price falls, but your dollar exposure does.
Raydium CPMM InitializeV2 is not deployed yet: check the chain, not the changelogevidence: B: mainnet RPC read + Raydium docs
- What we checked: reading each program's ProgramData account on mainnet and converting its last-deploy slot to a time (Oct 30 morning). Raydium CPMM was last deployed Oct 1, so the permissionless creator-fee pools in the InitializeV2 changelog entry (dated Oct 30) are not live yet. Raydium CLMM was last deployed Oct 22, matching its multi-pool upgrade. The pump.fun bonding curve and PumpSwap were both last deployed Oct 19. Meteora DLMM hasn't changed since June.
- Raydium's own entry says it was verified before deployment and to confirm the deployed program before relying on it. So the creator-fee cost (up to 1.495% per side, chosen by any pool creator) is coming, just not today.
- Lesson: a once-a-day, read-only check of the deploy slot for the venue programs you trade tells you the day a fee change goes live and the day a launchpad upgrade might break your transaction parsing. It's a handful of RPC calls per program.
pumpdata.fun: a free daily pump.fun stats file with wallet classes. Its top winners are not a copy listevidence: B: their recorder data; C: their classification
- What it is: an independent recorder of pump.fun curve and PumpSwap trades. It publishes a free daily JSON with totals, top coins, top deployers, failed-transaction share and top winners/exits, each tagged with a wallet class. A paid x402 lookup ($0.01 per wallet) returns a wallet's full record and class.
- In the Oct 29 file, 9 of the 10 top winners traded a single coin (one made +720 SOL from 178 trades on one mint), and the top exits sold tokens they never bought. The file itself says the winners and exits are the day's sellers, not a copy list.
- Other Oct 29 numbers: 70.6% of transactions failed (24.7M, costing 539.7 SOL in fees), 92% of launches had no trade in their second hour, and the top three deployers made 547 to 571 launches each with zero graduations.
- Lesson: a daily profit leaderboard is not a source of copyable wallets. Single-coin winners and wallets selling tokens they never bought should fail any copy vetting. A serial-deployer list (hundreds of launches, zero graduations) is a cheap skip filter.
Post-mortem: CATBOT, Oct 29's busiest coin. Curve skipped, tokens handed out, 99.95% off the topevidence: B: pumpdata.fun recorder
- What happened: CATBOT launched and graduated in the same second (Oct 28 20:47:10 UTC), going straight to PumpSwap. It was the most-traded coin of Oct 29: 48,152 SOL across 13,972 wallets, peak market cap $6.45M, close $2,964. The deployer bought only 0.05 SOL, and 114 wallets sold tokens they never bought, 5.32% of supply for 30.8 SOL. At least one trader on X called it bundled.
- Lesson: a launch that skips the curve and then tops the volume charts is a warning, not a signal. Wallets that sell without ever buying are being handed supply; drop them from any wallet scoring, and lean on liquidity, momentum and deceleration gates rather than volume rank.
Bloom/TradeWiz follow-up: drained addresses get drained againevidence: C: CryptoSlate, citing Bitquery
- What's new: Bloom says all nine affected users were refunded (9,408 SOL). CryptoSlate notes a 24-hour withdrawal-address lock does nothing against a copied key, because the attacker just signs a new transfer. In the TradeWiz case, two drained addresses received new deposits and were emptied again within 4 seconds. Half of the 48 Bloom wallets hadn't paid a Bloom fee in the past month.
- Lesson: once a key is suspected compromised, never top up or reuse that wallet. Move to a fresh key generated somewhere the old one never lived, and keep hot wallets at working balance only.
Sources
- https://financefeeds.com/binance-withdrawals-resume-48-hours/
- https://cryptocompass.com/articles/binance-withdrawals-after-the-hack-solana-and-xrp-paused-bitcoin-and-ethereum-open
- https://cryptoticker.io/en/solana-crash-below-100-binance-hack-sol-80/
- https://bitquery.io/investigations/binance-hack-who-profited
- https://docs.raydium.io/reference/changelog/2026-10-30-cpmm-initialize-v2.md
- https://docs.raydium.io/reference/changelog
- https://pumpdata.fun/report/2026-10-29
- https://pumpdata.fun/stats/2026-10-29.json
- https://pumpdata.fun/llms.txt
- https://cryptoslate.com/hackers-drain-9400-sol-from-bot-wallets-and-a-copied-key-is-blamed/
Evidence grades: A = large dataset, B = official docs, C = vendor/thread/small sample. Research notes for education only. Not financial advice.