Methodology
Data
- pump.fun bonding-curve trades and token creations, decoded from public Solana data on our own collector (exact SOL amounts).
- Public pump.fun home-feed callouts (handle, linked X handle, wallet, follower count, time).
- Transfers for traced wallets from public Solana RPC nodes (and a free-tier RPC key within a fixed budget).
- Exchange wallet labels: 421 hot wallets across 68 exchanges from free public sources (Dune Spellbook CEX address list, DefiLlama proof-of-reserves wallet lists, public community lists only where two or more independent lists agree), plus 6 deposit addresses inferred from our own traces. Coverage is partial; every label keeps its source.
Token patterns (score 0–100)
- Bundled launch: non-creator wallets buying in the creation slot and the next two slots (+15, up to +15 more by share of supply).
- Creator sell-off: creator sold ≥90% of its holding within 10 minutes (+20).
- Coordinated dump: ≥3 early wallets made their first sell within 10 seconds of each other (+20).
- Wash-like volume: ≥30% of volume from wallets that bought and sold back, with ≥5 SOL volume (+15).
- Serial group: creator belongs to a group with ≥5 launches in 7 days and ≥80% of them died (+20).
- Extraction: the token died (below 30% of peak, no trades for 2 h, not graduated) while the creator and creation-block buyers ended >0.5 SOL net positive (+10).
Wallet groups
- Creator wallets are linked when the same buyer wallet bought in the creation slot of ≥2 of their launches (generic sniper bots active with >10 creators are ignored), or when they were first funded by the same non-exchange wallet.
Exchange cash-out tracing
- From a creator wallet we follow plain SOL/USDC transfers (≥0.2 SOL or ≥20 USDC, swaps excluded) for up to 3 hops, keeping the 3 largest destinations per hop.
- A path ending at a labeled exchange hot wallet is reported; the address one hop before it is likely an exchange deposit address.
- Funding source: the first inbound SOL/USDC transfer to the wallet, and one more hop back.
Caller scorecards
- Only public pump.fun callouts are used; we show public handles and wallets, never real names.
- Bought before call, lead time, sold within 1 h of the call, token drawdown within 1 h, calls on serial-group tokens.
- Creator→caller transfers are reported only when found in traced creator wallets; a transfer is not proof of a paid promotion.
Case-file detectors
Case files (evidence packs) are built for the strongest flagged tokens, creator groups and callers. Each detector states only on-chain facts, gives a confidence from 0 to 1 that the pattern is present (high ≥0.75, medium ≥0.5), lists the transactions behind it, and makes no claim about anyone's intent or identity. A token case needs 2+ detectors, at least one of them other than creator sell-off at medium or high confidence.
| Detector | What it checks (facts) | False-positive risk |
|---|---|---|
| Bundled launch | Non-creator wallets buying in the creation slot and the next 2 slots; fires when 2+ of them, holding 3%+ of supply, are linked to the creator (in the creator's wallet group, or also early buyers on 2+ other launches by the same creator). | Sniper bots buy many launches; the grouping ignores same-block buyers seen on more than 10 creators' launches, but a bot that follows one prolific creator can still look linked. |
| Creator sell-off | Creator wallet sold 90%+ of its curve purchase within 10 minutes of creation; reports supply share, seconds to first sell and realized SOL. | Very common on pump.fun and often ordinary; never sufficient alone to open a case. |
| Wash-like trading | A wallet linked to the creator buys and sells back the same size (±2%) within 3 slots at least 3 times, or a linked wallet sells and another linked wallet buys the same size (±5%) in the same or next slot. | MEV/sandwich and arbitrage bots produce identical round trips; same-size hand-offs can be coincidence on busy tokens. Unlinked round trips are listed but do not fire. |
| Sells after a call | 2+ wallets linked to the creator sold 1+ SOL within 15 minutes after a public call, at least twice their selling in the 15 minutes before, and the market cap fell 30%+ within 1 h. | Many traders react to the same call; one bot operating several wallets for unrelated reasons looks the same. |
| Caller sold into the call | The caller wallet held the token at the call, then sold 50%+ within 30 minutes while other wallets were buying; reports exact minutes buy→call and call→sell, SOL in and out, and the buying by others in between. A caller case needs this on 3+ calls and 25%+ of its third-party calls. | pump.fun callouts are tied to a position, so buying before calling is built in; taking profit soon after posting can be ordinary trading. Only timing and amounts are reported. |
| Recurring launch-and-exit | A creator group with 5+ launches where 70%+ died (not migrated, no trades for 2 h, below 30% of peak or back under 40 SOL market cap) and either 30%+ had a fast creator sell-off or creators netted 5+ SOL. | Most pump.fun tokens die; this is weaker evidence than the per-token detectors and depends on the wallet grouping. |
Evidence packs and integrity
- Every pack lists supporting transaction signatures with Solscan links, UTC block times, SOL and token amounts; a timeline (creation, first buy per linked wallet, all creator trades, calls, peak, migration, last trade) with SOL market cap at each event; per-wallet P&L; and the funding/cash-out graph with each exchange label's source.
- Integrity: SHA-256 over the pack's canonical JSON (keys sorted, compact separators, UTF-8, without the integrity field), plus the build/retrieval time in UTC. The hashes of each build are kept in an append-only manifest. The PDF's own SHA-256 is returned in the X-PDF-SHA256 response header.
- Call links: the data source (pump.fun home-feed callouts) does not carry a tweet or Telegram URL. Packs link the caller's public X profile, the token page and the token's own listed socials, and the X post only where we captured it separately. Telegram calls are not collected.
Limitations
- Window: pump.fun tokens of the last 72 hours; tracing covers a capped number of wallets per run, on rate-limited public RPC.
- Bonding-curve trades only: P&L excludes fees, priority tips, DEX trading after migration and off-curve transfers; unrealized value at the last curve price is not achievable for large holdings.
- Some token creation times from the collector are a few seconds later than the first trade; packs order events by slot.
- Exchange labels are incomplete; many cash-outs go through unlabeled exchange addresses, bridges or swaps and are not shown.
- Patterns are statistical signals, and legitimate traders can match them.
Disputes and corrections
If you control a listed wallet or handle and believe an entry is wrong, email admin@deepcounterfeit.com with the address and a short explanation, or use the contact form. We review requests and correct or remove entries that are inaccurate. See also corrections & takedowns.
Automated on-chain pattern analysis of public blockchain data. Labels describe transaction patterns "consistent with" known manipulation techniques; they are not findings of wrongdoing by any person, and wallets are not linked to real identities. Exchange labels come from public community datasets and can be incomplete or wrong. Not investment, financial or legal advice. Disputes and corrections: admin@deepcounterfeit.com or /terms#corrections.