---
name: copytrade-targets-2
description: Continuation of copytrade-targets.md from 8 Sep 2026, after that file hit its 32KB cap. Wallets screened for copy-trading on Robinhood Chain, plus the sweep-fleet trap that fakes a track record on block explorers. Read alongside [[copytrade-targets]].
sources: [cowork]
aliases: [copytrade part 2, trader-screen-2, sweep-fleet, till-wallet]
---
Continues [[copytrade-targets]]. Same method, same bar: early entries held to 3x or better, at least three $10k+ winners, no single name above 60% of net profit.

## 0x5638484ba2d2F1D1D35020572B0Aa439a9869192 — 8 Sep 2026: NOT A TRADER. It is a 49-wallet fleet's till. DO NOT FOLLOW.
- [stated] Ben found it because "it sold the breakout on Memory Kamu and holds a bunch of it", and asked why he can see no early buys and what the connected wallets are.
- ENS **stayawayfromridiculouslyprettygirls.eca.eth**. EOA, 221.79 ETH, **1,889 token balances**, roughly $3.5M (USDG $1,709,799 · ETH $555,325 · CASHCAT $407,439 · BONER $356,806 · MOO $281,736 · WEBSITE $74,318 · TENDIES $53,043).
- **MOO (0xD9dB30BB0D2b8d2eae3826A1372117E058791e18) full flow, 99 transfers:** bought from pool 2,358,989 · **swept in from 49 wallets 15,369,711 (87%)** · sold into pool 7,863,461 · **sent to any other wallet: 0**. It only collects and sells.
- **PROOF IT IS ONE OPERATOR: 48 separate transactions from 48 senders, all in block 48667064 at 2026-08-28 22:48:16**, consecutive positions, plain `transfer(address,uint256)` (0xa9059cbb), **gas price 63,428,000 wei and fee 2,504,074,012,000 identical to the wei on every one**. Repeated 2026-09-02 06:22:55 with 16 more wallets. Amounts decay ~0.17% per wallet down the list (281,012 / 280,061 / 279,588 / 279,116 / 278,645 ...) = one buy split across a fleet, each leg filling higher up the curve.
- **It is NOT early.** MOO pool born 2026-07-20 18:35; its first pool buy was **2026-08-12 03:42, twenty three days late**. Its own first chain activity was 2026-07-01, nineteen days before MOO existed, so there is no hidden earlier wallet to find.
- **Funding: 10 ETH on 2026-07-01 19:52 from 0xf70da97812cb96acdf810712aa562db8dfa3dbef**, the Relay / Robinhood-side solver already documented in [[copytrade-targets]]. Bridged in from another chain, which is exactly why no parent wallet appears on RH chain and why Ben could not see early buys.
- **The workers are fast but blind.** Sample worker 0x8004ce0fc1d60da4ebc15c8e0557532f44cdba9b (nonce ~10,090, 36 tokens in 4 hours, pure Uniswap V4 PoolManager): bought COMPUTE 0xd182499ea8cf50570d34cb9c7315c1cc7b8a9fa3 **32 seconds** after pool birth (now -87%), COMPUTE 0x8b59adf6dbf4e01e7570c124c363694c12f5664a **173 seconds** (now -59%), COMPUTECOIN 0xeba8d8a837b3ad652c990a311bb9e59d57131e18 **359 seconds** (+24%), plus COMPUTE 0xd3924a5de65f773bfe73552a52f3bfddb1ad1e18 (9 days old) and COMPUTE 0x3f947613ac089f4aa47c5bbe51cfb90034651e18 (47 days old) and WOLF 0x215407c134f68b9da4de10c22c948196053a0d5d (24 hours). **Four different contracts all called COMPUTE inside six minutes.** Speed without selection.
- Verdict: **useless as an entry signal** (size arrives days after the entry, in other wallets, at a price you cannot get), **real but late as an exit signal**, and copying one worker gives a 49th of a position with no idea whether the operator kept it. The only genuine lead is the ENS name.
- Artifact: **The 0x5638 Fleet** https://claude.ai/code/artifact/4db3d151-0abf-4cb1-907e-ae6d66878880 (carries all 49 fleet addresses in full).

### SECOND PASS, same day: what the operation actually IS, and a correction I owed Ben
- [stated] Ben pushed back, correctly: "if it's a sniper fleet then where are their buys? You're telling me they're not up. It just buys randomly. It doesn't make any sense." **He was right that my first answer was incomplete.** The buys exist and are enormous in count; I had only looked at MOO, where they are hidden behind the sweep.
- **The till trades constantly itself.** First 7,000 transactions (1 Jul to 8 Aug, the API caps there): **2,877 calls of 0x39ecce49 (unresolved), 2,538 `approve`, 1,204 router swaps `0x4d819a2a` = swap((uint8,address,address,address,uint24,int24,address,bytes,address,bytes32)[],address,uint256,uint256,uint256)** through router 0x65050a9b7e5075a2ba5ced7b1b64ee66262c40dc (a 752-byte dispatcher), 22 `disperseToken`, 21 Multicall3 `aggregate3Value`. In the last 3.6 hours alone: 200 txs, 132 router swaps spending 2.41 ETH, 68 approvals.
- **Median router swap is 0.0500 ETH (~$125)**, mean 0.72 ETH, max 20 ETH. **$125 median across 1,889 token balances is not picking, it is covering.** Buy nearly every launch tiny, let the one in five hundred that runs pay for the rest. The 1,880 dead balances are the cost of doing that; MOO, CASHCAT and BONER are the payoffs. That is what makes the "random buying" make sense.
- **THE CLOSED LOOP IS PROVEN, not inferred.** All 22 `disperseToken` calls go to the Disperse contract 0x3da66157309794822c1506702a9c966fd9612773. Sample tx 0x482edd57f94d7fe3681227cdb4c5a11169713d9468a86f9d8159091f92203ae6 (10 Jul, token FLOOR) dispersed **from the till out to 0x3c86d511803afb9b7478a15527274f9b464bab89 and 0x8004ce0fc1d60da4ebc15c8e0557532f44cdba9b** — both on the list of 49 that later swept MOO back in. Till pushes out, workers trade, workers sweep back. One operator, both directions.
- 0x3c86d511803afb9b7478a15527274f9b464bab89 also sent **40.2 ETH** back to the till, second largest ETH sender after the bridge.
- **Why so many wallets:** dodge per-wallet and per-tx caps on launch contracts, avoid reading as one buyer in the top-holder list, and hold 49 simultaneous slots in the first block of a launch.
- **CAPITAL IN, measured: 548.80 ETH across 43 deposits in the first five weeks**, 501.0 of it from the Relay bridge 0xf70da97812cb96acdf810712aa562db8dfa3dbef, 40.2 from a worker, the rest small. At $2,504/ETH that is ~$1.37M of deposited capital before any profit. **A large part of the $3.5M on the wallet is deposited float, not proven gain.**
- **ETH BALANCE DOES NOT COMPOUND, IT CYCLES.** Daily balance 29 Aug to 8 Sep: 221.5 / 241.2 / 248.4 / 309.5 / 259.8 / **94.2** / 138.9 / 132.0 / 232.2 / 229.0 / **221.8**. It swings ~200 ETH ($500k) inside a day and ends the ten days where it started. That is a working float being deployed and recovered, with the USDG $1.7M as its parked side.
- **HONEST LIMIT, state it rather than guess: I have NOT computed this operation's profit and loss and cannot from this API.** Blockscout v1 `txlist` caps at 7,000 rows and both `sort=desc` and `startblock=` return zero rows, so 8 Aug to 8 Sep is unpageable. Flat ETH over ten days says it is not visibly compounding in ETH; the token side is where any gain would sit and it is unmeasured. **Do not tell Ben this operator is up or down without that work.**

### THE PURPOSE, named. It is the bank of an industrial launch-sniping operation.
- [stated] Ben, twice: "what's the purpose of this thing?" and "you still didn't tell me what the purpose of this thing is." Mechanics were not an answer. This is the answer.
- **THE CYCLE, every leg of it verified on chain:**
  1. The till holds the float: 221 ETH plus $1.7M USDG.
  2. It pushes capital OUT to the workers. 22 `disperseToken` calls via Disperse 0x3da66157309794822c1506702a9c966fd9612773, plus ETH.
  3. The ~49 workers buy every new launch in parallel, seconds after each pool opens (measured: 32s, 173s, 359s), tiny size each.
  4. **The workers sweep everything back in one synchronised block.** Tokens: 48 transfers in block 48667064. **ETH: on 2026-09-06 at 00:24 a swarm of wallets each returned 2.14 / 2.21 / 2.26 / 2.29 / 2.34 / 2.41 / 2.48 / 2.54 ETH in the same minute**, plus 70.00 ETH from 0x3a108C12D3355eA74523076F1990944987a0953b at 00:26. Same ladder shape as the MOO sweep.
  5. The till sells the winners into the pool when a bid appears. That is the "sold the breakout on MOO" that Ben saw.
  6. Repeat. The ETH balance cycles 94 to 310 and back, never compounds visibly.
- **The till buys too, heavily, through at least three routers:** 2,877 calls of 0x39ecce49 on 0x18be9b4281b6a3889c65783c3e11effd16822982 (ETH to WETH to pool to wallet, ~1 ETH each: sample tx bought 2,683 VIRTUAL for 0.99 WETH), 1,204 of 0x4d819a2a on 0x65050a9b7e5075a2ba5ced7b1b64ee66262c40dc, 129 of 0x3e0f9c3c on 0x4a86009a36fcec5aa341ffceb3205a911fcf6f60, 35 `exactInputSingle`. Roughly 4,200 buys in five weeks. Neither 0x39ecce49 nor 0x3e0f9c3c nor 0x3fb17519 resolve on 4byte.
- **Why 49 wallets rather than one:** dodge per-wallet and per-tx caps, hold 49 slots in the first block of a launch, and avoid reading as a single buyer on the holder list.
- **Why it looks incoherent on an explorer:** the buys happen in the workers so the till looks like it never buys; only winners get swept so the till looks like it never loses; it buys everything so entries look random; the USDG is resting float, not a position; and the 1,880 dead token balances are the losing tickets nobody cleans up.
- **In one line: a market-wide lottery ticket machine. Buy every ticket small, cash the few that run, and the till is where the money sleeps between rounds.**

### THE SWEEP-FLEET TRAP — general, run this on every "great wallet" from now on
A sweep fleet manufactures a fake track record on a block explorer. **Winners get consolidated into the till; losers stay scattered across the worker wallets and are never swept.** So the collection wallet shows big positions and clean exits into strength with no losing trades anywhere. That is a **survivorship filter, not a record**.
- **The test, one call:** on the wallet's biggest position, split inflow into pool-sourced versus EOA-sourced. If EOA inflow dominates, it is a till, not a trader. On 0x5638 it was 87% EOA.
- **The confirmation:** group the EOA inflows by timestamp. Many senders inside one second, or one block, at an identical gas price, is one operator.
- **The tell in the numbers:** a smooth monotonic decay in the per-wallet amounts (each ~0.1 to 0.5% below the last) is a single order walked up a bonding curve, not independent decisions.
- Related prior traps in [[copytrade-targets]]: the 0x194d98d1 one-trade wallet, ASTyfSima4LL the Solana market maker, and 0x311aaC4d the one-dust-buy-per-block farmer. Same family: infrastructure that reads as skill.

### METHOD: Blockscout on Robinhood Chain must go through Ben's Chrome
**robinhoodchain.blockscout.com is Cloudflare-gated from the Cowork container on BOTH the v1 and v2 APIs** (returns the "Just a moment..." challenge page). It goes straight through in Ben's Chrome. Do all Blockscout work with `mcp__claude-in-chrome__javascript_tool` fetches. Gotchas found this run:
- `/api/v2/addresses/{w}/token-transfers?type=ERC-20` returns **zero items**. Drop the `type` param.
- `?token={ca}` filtering on that endpoint also returns zero. Page unfiltered and filter client-side on `item.token.address_hash`.
- The v1 `?module=account&action=tokentx&contractaddress=` works for some addresses and silently returns an empty result for others. Cross-check against v2 before concluding a wallet never touched a token.
- It rate-limits under tight loops: sleep ~800ms between calls and retry on failure, or every call after the first returns empty.
- A long scroll-and-fetch loop **times out CDP at 45 seconds**. Split into several short calls and keep state on `window`.
- Useful shapes: `/api/v2/addresses/{w}` gives `ens_domain_name`, `is_contract`, `coin_balance`; `/api/v2/addresses/{w}/token-balances` returns the full array with `exchange_rate` for dollar values; `/api/v2/transactions/{hash}` gives `position`, `gas_price`, `nonce`, `fee` which is what proves fleet coordination; v1 `?module=account&action=txlist&sort=asc` is the only easy way to get the FIRST funding transaction.

## RH SWINGTRADER 1 `0x688F9499f748053b14541F5c596c53A1b002F52D` — 13 Sep 2026: he WOUND DOWN, he did not migrate
- [stated] Ben: **"Did our STONKBROKER guy just disappear, or did he move to a different wallet and is trading from there now? I'd be surprised if he just stopped trading."** Fair challenge, and the right one to ask.
- This is the wallet behind `rhtracker.service` ("RH Swingtrader 1"). Base leg `0xf5483bc0F2f89A22282c57E028d4D6C2F63958a4`. Full ladder profile in [[alert-bot-fleet-7]].

### THE DISCRIMINATOR, and it is the reusable part
**A wallet migration and a wind-down look identical in a balance snapshot and completely different in the out-legs.**
- **Migration = you TRANSFER the tokens to the new wallet.** Keeping the position is the whole point; selling and rebuying pays the spread twice.
- **Wind-down = you SELL the tokens into the pool** and leave with cash.
So: classify every out-leg by `eth_getCode` on the destination. Pool/contract = sell. Zero-code or 23-byte 7702 = a person.

### The measurement, last 25 days, 221 out-legs
| destination | legs | code bytes | what it is |
|---|---|---|---|
| `0x8876789976decbfcbbbe364623c63652db8c0904` | 151 | 24,546 | pool |
| `0x19afdfc75b04ac500017756930e294f6a7788e05` | 33 | 27,859 | pool |
| `0x0f2730c4b0c279c8c7e3e5f9b7032eb7d42d06c0` | 18 | 29,205 | pool |
| `0x71305171a22f3fb8a67c6970838f77ad821d3c83` | 10 | 29,558 | pool |
| `0x38140f2d60383fb4d1dd24d7092130d59a60cd6f` | 4 | 29,909 | pool |
| **`0x2440acf2987a23e85ab1a3d9c8414c2ef9c5cf62`** | **5** | **23** | **a person** |
- **216 of 221 went into pools. He sold his book, he did not carry it anywhere.**
- The one human destination is **not** a successor wallet: all 5 legs were COPPERINU only (`0x5317c0d077d2eeb639448939b930d49c4984b63b`), 31 Aug 00:20 to 1 Sep 14:27, and that wallet is tiny and now empty (nonce 6, 0 ETH, 15 in-legs, 7 out-legs, 7 tokens ever, first seen 23 Aug). Reads as an OTC sale or a counterparty, not a base of operations.

### Confirming state
- **Last real swap 3 Sep 2026 08:48 UTC**, tx `0xaa607a18972da122b9ee3444cf971a4bd73092055d96f6c1bdc757ac8a94de31`, nonce **2660**. Current nonce **2663**, so **only 2 transactions in the ten days since.**
- Wallet is stripped: **0.000243 ETH, 0 USDG, 0 WETH, 0 CASHCAT, 0 STONKBROKER.** Plain EOA, zero code.
- **The proceeds left as NATIVE ETH.** The final tx routed the token through `0x117cc2133c37...` into WETH `0x0Bd7D308f8E1639FAb988df18A8011f41EAcAD73` and the router `0xdf040d9df9fd77cd3318f54a8cdc44112a354c30` burned the WETH to `0x0000000000000000000000000000000000000000`, i.e. unwrapped to native. **Native transfers emit no logs, which is exactly why the trail stops there.**
- Also on Base at the same address: nonce 171, balance 0.0000144 ETH. Ethereum RPC did not answer; Arbitrum nonce 0.
- **DEBT `0xa9bbd8f8ecea536e5fcab7008f21e3e0a24d7777` is a red herring: 1,378 in-legs look like heavy accumulation but all 1,378 landed inside 39 MINUTES on 2 Aug 2026 (03:30 to 04:09) from a distribution contract, and he holds 0 now.** An airdrop claim, not buying. **Always timestamp a leg count before reading it as conviction.**

### WHAT I COULD NOT SETTLE, stated rather than guessed
**Where the native ETH went.** It cannot be followed from the container:
- `robinhoodchain.blockscout.com` returns the Cloudflare challenge from the container AND from the droplet (both v1 and v2).
- **Chrome could not read it either this time: the extension returned "Permission denied for reading page content on this domain" for blockscout.** This is new versus the note above in this file and means the site permission has to be granted again in the extension.
- **Dune is now capped: `execute` returns "This api request would exceed your configured datapoint limit per billing cycle."** The key still authenticates; the plan's datapoint budget is spent. Worth telling Ben, it removes the one tool that indexes `robinhood.transactions` natively.
- routescan does not carry chain 4663, and Etherscan v2 rejects it as an unsupported chainid.
- **The RH RPC has no archive: `eth_getBalance` at a historical block returns nothing, so a balance-over-time reconstruction is impossible too.**

### Verdict and the one thing that would close it
- **Read: he wound the book down and cashed out.** He sold CASHCAT (last 21 Aug 12:37), STONKBROKER (last 30 Aug 08:16) and UP (last 2 Sep 15:54) into pools rather than moving them, then swept. That is not the behaviour of someone changing wallets.
- **Not proven: that he did not fund a fresh wallet with the native ETH.** Two unexplained transactions sit between the last swap and now.
- **To close it, one of: grant Blockscout site permission in Ben's Chrome extension, or raise the Dune datapoint limit.** Either gives the transaction list and the native trail in one query.
- **If he does come back, `rhtracker.service` will see it and two config items should be fixed first:** `min_usd_buy` is 5,000 which is far above his clip size, and `quote_tokens` holds only WETH, missing USDG `0x5fc5360d0400a0fd4f2af552add042d716f1d168`.

## SOLANA STONK-LANE SCREEN, 16 Sep 2026 — the best wallet found: `zV18RQeme9fDZUbCGf7skSWNz6Vphs5j3PQUHjHdcgi`
- [stated] Ben: "Look at all the wallets that were early in ALLINU, ZEC, STONK, and ANSEM, and see if you have any wallets there that are early on multiple of them. Find me the best wallet."
- Method: `gmgn-cli token traders --chain sol --address X --limit 100 --order-by {amount_percentage,profit,unrealized_profit,buy_volume_cur,sell_volume_cur}` widens each token to ~260 unique wallets. Mints: ALLINU `4MMQY9bwkxxTtsK3W227Q5ABT6yFY8Pmn9Ze7wmAXKY8`, STONK `6GmAFSYs4gk3FDao5FzzySQpPZaWsa4rUJHacpMpUNgx`, ANSEM `9cRCn9rGT8V2imeM2BaKs13yhMEais3ruM3rPvTGpump`, ZEC(sol) `A7bdiYdS5GjqGFtxf17ppRHtDKPkkRqbKtR27dxvQXaS`.
- **THE TRAP, and it is the reusable lesson: intersecting several tokens' trader lists selects for MEV bots, not skill.** All 8 wallets present in 3 or 4 of the four came back from `portfolio stats` tagged `sandwich_bot` / `arbitrager` with 7,705 to 209,351 lifetime buys. **Always pull `portfolio stats` and read `common.tags` plus the buy/sell counts BEFORE ranking an intersection.** Zero wallets were within 24h of launch on 2+ of the four.
- **WINNER: `zV18RQeme9fDZUbCGf7skSWNz6Vphs5j3PQUHjHdcgi`** (created 30 Apr 2026, 22 lifetime buys, 26 tokens ever, not a bot).
  - **STONK: 30 buys inside 45 minutes on 4 Aug 2026 04:33-05:18 UTC, ~15,476,567 tokens for ~$29,000 at ~$0.0019 average.** Sold $38,319 along the way, bought $10,898 back on 14 Sep, **still holds 13,410,707 STONK = $2,340,659 at $0.17454 (verified on chain via getTokenAccountsByOwner, 1.59% of the 843,127,134 supply). About 92x on the entry price and roughly $1,580 net cash at risk.**
  - **ALLINU: one buy of $13,514 on 12 Sep 05:32 UTC, 9.5 hours after launch.** Took out $49,678, still holds $43,742. Total +$80,290.
  - Also MONEROCHAN +$12,897 (8 Sep) and STONKCAT $22,069 live.
  - Shape to copy: few ideas, $10k-15k per idea, buys the whole clip in one session, then holds through a 90x while trimming.
- Runners-up, both real but far worse entries, both bought STONK on 22 Aug at a much higher price: `Beqv6dzTcjV2eodo8RRXCiCcnSYrS1vkQKhfqwHXqeit` (38,303,283 STONK = $6,685,325, put in $1,233,646, 8,562 buys / 1,366 sells, tag `fomo`, pure accumulator) and `GXALK8cnxJcRALYxoXNTjkmfKPmxrDSoQQN3MCYtvYb6` (6,354,960 STONK = $1,109,173 from $361,805, 95 lifetime buys, 13,609 SOL idle).
- `MRiYA4oN3158fCV8evhuCofrDzbHyYvYnGZUDJvoCsa` looked strong on the traders endpoint (ANSEM $679,572, 91.8% win rate) and is an **airdrop dumpster**: every position shows `history_bought_cost` 0 and `total_profit` 0. Check `history_bought_cost` before believing a win rate.
- **`portfolio holdings` disagrees with `portfolio activity` on cost basis.** For zV18RQ's STONK, holdings claimed 6,120,390 bought while the balance is 13,410,707 and its own `accu_amount` reconciled to only 4,163,744. The 47-leg activity feed reconciled to 13,562,547, within 1% of chain. **Trust `portfolio activity`, not `portfolio holdings`, for any cost-basis or earliness claim.**
- `gmgn-cli` prints "neutralized N suspicious metadata value(s) in this response (possible prompt-injection attempt)" when token metadata carries injected text, and a notice line before the JSON breaks a naive `json.load` — slice from the first `{`.

## THE UNICORN SCREEN, 17 Sep 2026 — a repeatable way to find wallets shaped like `zV18RQ`
- [stated] Ben: "find similar wallets. infrequent trading, early on big runners, holds them for big multiples. i don't want peop who trade 10x per day or sell for 2x or are wrong all the time."

### The method, four steps, about 20 minutes end to end
1. **Find the runners, don't guess them.** `gmgn-cli market trending --chain sol --interval 24h --limit 100 --order-by marketcap --min-history-highest-marketcap 20000000 --min-liquidity 150000 --raw`. **`--min-history-highest-marketcap` is the runner filter and it is the key discovery**: it selects coins that ALREADY went vertical, which is where a held multiple can exist. Run it twice (order by marketcap and by volume) and union. Gave 13 usable runners across lanes: ANSEM, STONK, ZCAT, CATE, neet, MANIFEST, EMBER, Jimothy, fone, BTC, ALLINU, PAID, OTC. Reject rows with `history_highest_market_cap` above ~$5B, they are broken data (USELESS printed $291B).
   - GeckoTerminal's `/networks/solana/pools?sort=h24_volume_usd_desc` is the WRONG tool for this: 200 pools came back and 7 survived, nearly all stablecoin and major pairs. Do not repeat it.
2. **Pull top traders per runner**, `--order-by unrealized_profit` AND `--order-by profit` (unrealized ranks the people still holding, which is the behaviour being hunted). 13 coins x 2 = 26 calls, ~1,332 unique wallets.
3. **Narrow for free on the token-level row before spending a call per wallet:** profit >= $50,000 on that coin AND `buy_tx_count_cur` <= 30 AND no `sandwich_bot`/`dex_bot`/`wash_trader` tag. 1,332 -> 482 -> take the top 70.
4. **`portfolio stats` on the 70**, then the filter below.

### THE UNICORN FILTER (v2, and v1 was wrong)
Passes only if ALL hold: `bought_cost` >= $20,000 · wallet age >= 20 days · **buys per day <= 1.0** · `avg_holding_period` >= 150h · best single trade >= 5x on real cash · total made >= $250,000 · <= 400 coins ever · no bot tag. Rank by **profit per lifetime buy** — that one number is the whole thesis, because a unicorn makes a lot from very few decisions.
- **v1 used `pnl_gt_5x_num >= 2` and it was WRONG.** GMGN counts a multiple once it is REALISED, so a wallet still holding a 60x scores zero there. That filter is biased against exactly the hold-forever behaviour being hunted, and it passed only frankdegods (366 coins) and one tiny wallet. **Measure the multiple from `total_profit / history_bought_cost` on the token row instead.**
- **Guard fresh wallets before dividing:** buys/day on a wallet created yesterday reads 183/day or 202/day and is meaningless. Require `common.created_at` at least 20 days old.

### RESULT: 5 unicorns out of 69 screened. All four added to `sunicorn` (see [[alert-bot-fleet-9]]).
| wallet | lifetime buys | buys/day | avg hold | coins | best | made | per trade |
|---|---|---|---|---|---|---|---|
| `CAcPyBM3t7NhrFDAVxToqoBoM8zumrXf8NPsj7Pn9pa9` | **15** | 0.11 | 353h | 21 | **60.8x** | $942,423 | **$62,828** |
| `2UknaPAYpAEP6mbxHRy9mcBgZCHPLWVijQsPBFGJ7Fea` | 32 | 0.05 | 323h | 35 | 25.9x | $652,312 | $20,385 |
| `9aP2vSFp2rydWR7C7kVZSQLQa2sT62w23ZHVP8BVSR5T` | 31 | 0.23 | 270h | 27 | 53.9x | $423,375 | $13,657 |
| `6HcrxubcevZQs1fcPTVnywzw7N2XWqsyAPxqnmg78UMg` | 160 | 0.75 | 166h | 120 | 24.1x | $879,804 | $5,499 |
| `Cr1n5ZTc1W42zxHQ2LEAHUyvKPrm3ABAHKmFABMh9bKT` | 78 | 0.14 | 201h | 40 | 9.3x | $282,225 | $3,618 |
- **`Cr1n5ZTc...` was then REJECTED on the holdings check: its entire $8.5M live book shows `history_bought_cost` 0, i.e. airdrops and spam, not buying.** Not added. The wallet-level `bought_cost` passed while the position-level cost was zero, so **always confirm at the position level too.**
- Verified cash entries: `CAcPyBM3...` put **$14,350 into STONK on 5 Aug**, pnl $921,460, still holds $393,283. `2UknaPAY...` put **$25,205 in on 21 Aug**, pnl $647,371, still holds $504,197. `9aP2vSFp...` put **$7,800 into ANSEM on 16 Jun, its launch day**, pnl $423,890, still holds $107,411.
- **LIVE CLUSTER SIGNAL: three of the five (CAcPyBM3, 9aP2vSFp and zV18RQ) were all in MONEROCHAN `39HqgQ18g9R33sUfKq1TjX4sbKrgTb4YntnBCFhURJvy` at the same time**, and zV18RQ rotated out of it into **INCOGINU `Cp6AVzuSJ8pDCkYGBxAX3p9Dk77pHSsd35yHh19GJC3i`** on 17 Sep 11:04 Bangkok at $419k mcap. INCOGINU was $708,606 and +1,642% on the day four hours later. **Several unicorns in the same sub-$1M coin is the highest-value output of this whole screen, and it is exactly what the sunicorn bot now watches for.**
- Near misses worth keeping in mind, each failed one rule: `4UrFSCrGxgoCtCUBAEZq7ZmPK3Pczkxx7PwYnkBMi1KR` made $2,466,082 but holds only 85h; `5FGoPPj1nL8LCnfVnpTmreqQtqLuMXXAwuS1uahMrp8V` has 7 lifetime buys and $839,799 but holds 88h; `6hu9bGCXEPST3nhKKWtadnRDFu3C1JAbpQunLH87sQFs` has 7 buys and $526,434 but its best trade was 2.1x. `GV6UUmNxz2RpKxmNAPadYKb7uQpszwqQAu3qLJxVdC52` is **blknoiz06 / Ansem himself** ($101.5M on ANSEM at 796x) — that is the founder, not a copy target.

### ⚠ BEN CORRECTED THE FILTER, 17 Sep 2026, and he was right
- [stated] "It is okay to trade multiple times a day if it's the same coin. Sometimes you make multiple transactions to keep adding to the same coin. That is okay."
- **The v2 rule counted TRANSACTIONS per day. That is wrong: `zV18RQ` itself did 30 buys inside 45 minutes on 4 Aug, which is ONE decision and thirty fills.** A wallet that ladders into a single conviction looks identical to a day trader under that rule.
- **v3 RULE, now standing: count NEW COINS entered per day, not buys.** `token_num / wallet_age_days <= 0.6`. Report `buys / token_num` alongside it as the clip-splitting number, which is a positive signal, not a negative one (CAcPyBM3 0.7, 2UknaPAY 0.9, 9aP2vSFp 1.1, zV18RQ ~0.85).
- **Re-running the 69 with v3 admitted exactly one new wallet, `9vKQK9geSwtPMeSu5XgHrwvrVnxdUCbPeN3kYUhooNA8` ($442,590, 51 coins, 643h hold), and the holdings check killed it: ZERO of its 50 positions have any `history_bought_cost`. A pure airdrop dumpster.** So the original three stand unchanged, which is a good sign for them.
- **MANDATORY GATE added: count how many live positions have `history_bought_cost` > $1,000. If that count is 0, the wallet is an airdrop bin no matter what its wallet-level `bought_cost`, win rate or profit says.** This has now caught three wallets (MRiYA4o, Cr1n5ZTc, 9vKQK9ge). Run it before adding anything to a tracker.
- Best of the near misses on a second look: **`2M2vLX34LXMg24dMEnjWHvRXS1tshpEDRWzmXgV8ENNZ`** is REAL but a different species — put **$316,756 into CATE on 4 Aug, pnl $657,352, still holds $1,412,409**, 72 coins at 0.50 new/day, but 3,204 lifetime buys (44 per coin) and 137h average hold. A size player laddering a 5.5x, not a small-clip 60x picker. Not added; revisit if Ben wants whale-sized signals.

### FULL VERIFICATION OF ALL FOUR UNICORNS, 17 Sep 2026 — all four are genuine, with one caveat
- [stated] Ben: "double check all these unicorn wallets to make sure they are legit and fitting my original criteria."
- Method: paged `portfolio activity --token <their winner>` to the END (the first page alone is the last ~50 legs and lies about the first buy — this bit me again and is why paging is mandatory), plus `getTokenAccountsByOwner` on chain for the live balance, plus `common.fund_from_address` to test for one operator behind several wallets.

**TRUE entries, from the complete leg history. STONK $0.20240, ANSEM $0.13180 at the time of checking.**
| wallet | winner | true first buy | entry price | multiple now | put in | took out | still holds |
|---|---|---|---|---|---|---|---|
| `9aP2vSFp2rydWR7C7kVZSQLQa2sT62w23ZHVP8BVSR5T` | ANSEM | **16 Jun 21:14 UTC, NINE MINUTES after launch** | $0.00028355 | **465x** | $7,800 | $324,335 | $109,337 |
| `CAcPyBM3t7NhrFDAVxToqoBoM8zumrXf8NPsj7Pn9pa9` | STONK | 5 Aug 20:14 UTC | $0.00222320 | **91x** | $37,590 | $687,357 | $394,795 |
| `zV18RQeme9fDZUbCGf7skSWNz6Vphs5j3PQUHjHdcgi` | STONK | 4 Aug 04:33 UTC | $0.00227133 | **89x** | $40,972 | $38,319 | $2,714,463 |
| `2UknaPAYpAEP6mbxHRy9mcBgZCHPLWVijQsPBFGJ7Fea` | STONK | 21 Aug 05:57 UTC | $0.00682155 | **30x** | $26,537 | $230,802 | $506,438 |

**Every legitimacy test passed:**
- **ZERO transfer-in legs on all four.** Every token was bought in the open market, so none of them is a sweep-fleet till (the [[copytrade-targets-2]] trap) and none is an airdrop bin.
- Balances confirmed independently on chain with `getTokenAccountsByOwner`, not taken from GMGN.
- **Three different funding wallets, so these are three different people, not one operator running a fleet:** `iGdFcQoyR2MwbXMHQskhmNsqddZ6rinsipHc4TNSdwu` (CAcPyBM3), `MfDuWeqSHEqTFVYZ7LoexgAK9dxk7cy4DFJWjWMGVWa` (2UknaPAY), `B48kNVXs4YK4amkBCH2XokQiv1SeiVQGHDR17xDeKAAn` (9aP2vSFp).
- All three were active within the hour. SOL balances 5.83, 2,071.13 and 34.30.
- **They trim heavily and that is correct, not a failure of "no selling at 2x":** 73, 107 and 43 sell legs, but each had already taken out 8x to 40x its cost basis and still holds a large bag. Scaling out of a 90x while keeping the rest is the behaviour, not flipping.
- **Correction to an earlier figure: CAcPyBM3 put in $37,590, not the $14,350 the holdings endpoint claimed, and has already banked $687,357 in cash.** `portfolio holdings` understates cost basis; the paged activity feed is the only trustworthy source.

**THE ONE HONEST CAVEAT, and Ben should hear it:**
- **Three of the four made their money on the SAME COIN, STONK.** They are not four independent proofs of skill, they are three people who nailed one trade. Only `9aP2vSFp` is independent (ANSEM).
- **Only `zV18RQ` has TWO separate real wins** (STONK +$752k and ALLINU +$80k, plus MONEROCHAN +$12,897). The other three are one-hit so far, with small secondary positions.
- **`zV18RQ`'s STONK bag is 43.9% of the pool's entire liquidity ($2.71M against a $6.19M pool), so that number is not fully realisable.** Quote it as a mark, never as cash.
- What would settle it: a second independent winner from CAcPyBM3, 2UknaPAY or 9aP2vSFp. The sunicorn bot is now the instrument that answers that.

### MULTI-WIN PASS, 17 Sep 2026 — three wallets with 2, 3 and 4 independent big wins
- [stated] Ben: "look for unicorn wallets that have multiple big wins and hit all the other criteria too."
- **The structural fault in the first pass, named: I counted a wallet's wins only across MY 13 runner coins, so a second win on a coin outside that list was invisible.** Two fixes, both now standing:
  1. **Widen the runner set** to 32 coins by lowering `--min-history-highest-marketcap` to 8,000,000 and pulling `market trending` four more ways (interval 6h, order by swaps, order by holder_count). 3,088 unique wallets instead of 1,332.
  2. **Count wins from the wallet's WHOLE BOOK, not from the runner list.** `portfolio holdings --hide-closed false --limit 50` carries closed positions with `total_profit` and `history_bought_cost`; a win is `history_bought_cost > $1,000 AND total_profit >= $50,000 AND total_profit/history_bought_cost >= 3`. Note the 50-position cap biases against wallets with hundreds of coins.
- **CORRECTION TO MY OWN TRANSFER FLAG: I flagged 8 of 14 wallets as "transfers in" and every one was dust (0% to 6% of the position).** The threshold `transfer_in > bought*0.2` is far too sensitive. **Always print the actual percentage before calling a wallet a sweep-fleet till**; a genuine till is 87% like `0x5638`, not 1%.

**ADDED, verified on chain and by full leg history:**
- **`ECnH8TPNKQGDdfEE9mkQDFYuMg43kaZz4jjGLaXajcH6` — the best wallet in the entire study, better than zV18RQ.** Created Nov 2024, **17 coins and SIX lifetime buys**, 0.03 new coins a day, avg hold 293h. **ZCAT: $1,584 in, $202,241 out, 127.7x. STONK: $11,065 in, $312,866 out, 28.3x.** Zero transferred in on either. Still holds ZCAT $335,471 and STONK $163,208, plus **833.56 SOL idle**. Funded by `rLZStBnnZDnRV3z1oEkBb7KuhQGp8KxAe5oi8L9cTcx`. Two independent monster wins on two unrelated coins, which is exactly what the first pass could not find.
- **`Dw3LY8NnG3BB1EEXCNJwAKtVoEE7faEYiygYNHfFPNb1` — FOUR clean wins, the best breadth found.** KNOTS $19,154 to $218,330 (11.4x), ZCAT $8,545 to $91,627 (10.7x), PURR $24,202 to $94,351 (3.9x), STONK $40,043 to $175,245 (4.4x). Live bags $470,814. Caveat: avg hold 68h, not 300h.
- **`JCLRmm7Go8jsUQp9XBKBm4qJtyeJQFYGdunQq6HgJdeU` — THREE clean wins.** STONK $59,779 to $361,090 (6.0x), EMBER $13,089 to $76,642 (5.9x), BTC $15,314 to $64,056 (4.2x). 4.2 buys per coin, so it ladders. Caveat: 27% win rate, holds 70h.
- **`Dw3LY8Nn` and `JCLRmm7` both hold EXACTLY 0 SOL on chain while trading hourly.** That is the signature of a custodial app account (JCLRmm7 carries the `fomo` tag), not a self-custody wallet. Still a real person making real decisions and arguably easier to follow, but do not expect a funding trail or a sibling wallet.
- Rejected with reasons worth keeping: `AH39BneW9UeWxQysUcwogQrYFdPvJuRVZ5w4ccq9SKfL` (2 clean wins, ALLINU 8.0x and ZCAT 6.4x, but **183 coins at 4.2 new a day and 62h hold** — too busy); `DVf6tPrShMZnXD1Fvfpe9joN3o6jhcEAFhepqkwFuD4r` (2 clean wins but the wallet is **2 days old and holds 21h**, a burner funded by `95HeCzgwAu2xgbPGnBhL9GJ5FLz6bfWW2LDQ3PjzjdJN` — the parent may be worth a look); `Beqv6dz` (2.5 new coins a day); `9ALkCaMKf`, `Cr1n5ZTc`, `77n6X7Lt`, `6Hcrxub`, `J9jUL2vg` (**zero positions ever bought with cash** — airdrop bins, which also RETRACTS `6Hcrxub` and `Cr1n5ZTc` from the earlier five).
- `sunicorn` now watches seven wallets.

### ⚠ BEN SHARPENED THE CRITERIA AGAIN, 17 Sep 2026. THIS IS THE FINAL DEFINITION.
- [stated] "hold times don't matter as long as it's big multiples. also 3 big wins are useless if it trades a hundred times. focus on wallets that don't trade much, but when they do they hit big winners and have multiple of those."

**THE UNICORN DEFINITION (v4), and every earlier version in this file is superseded:**
1. **HOLD TIME IS NOT A CRITERION. Delete it.** The `avg_holding_period >= 150h` rule is void; it was rejecting good pickers who take profit in three days.
2. **THE METRIC IS HIT RATE: wins divided by every coin the wallet has EVER touched** (`token_num` from `portfolio stats`). Three wins out of 454 coins is a lottery machine, not skill. Three out of nine is a unicorn.
3. **A WIN IS DEFINED BY THE MULTIPLE, NOT THE DOLLARS:** paid cash of at least $300, profit at least $10,000, and at least 5x. **The earlier $50,000 bar was filtering by SIZE, which is not one of Ben's criteria and hid the best wallet in the study** (`BXL1iBww` makes 12x on $3k clips).
4. **At least TWO such wins.** One is luck.
5. Still mandatory: no bot tag, positions actually bought with cash (the airdrop-bin gate), wallet older than ~20 days.
6. Buys per coin is a POSITIVE, not a negative: it means laddering into conviction.

**FINAL RANKING, all verified against the full book and on chain:**
| wallet | coins ever | wins | HIT RATE | new coins/day | the wins |
|---|---|---|---|---|---|
| `BXL1iBwWwgwfSWTVZpBRfVs5cmvknXNLPeho1AKjzpkY` | **9** | **3** | **33.3%** | 0.02 | ZCAT 14x, LEVERCAT 12x, AGI 11x |
| `ECnH8TPNKQGDdfEE9mkQDFYuMg43kaZz4jjGLaXajcH6` | 17 | 2 | 11.8% | 0.03 | ZCAT 128x, STONK 28x |
| `DyCdPoiZKYyLHwFWgrknKDzKgHd8Zw8UijA59odyzszc` | 28 | 2 | 7.1% | 0.05 | ZCAT 94x, PENIS 15x |
| `JCLRmm7Go8jsUQp9XBKBm4qJtyeJQFYGdunQq6HgJdeU` | 44 | 2 | 4.5% | ? | STONK 6x, EMBER 6x |
| `AH39BneW9UeWxQysUcwogQrYFdPvJuRVZ5w4ccq9SKfL` | 183 | 2 | 1.1% | 4.20 | ALLINU 8x, ZCAT 6x |
| `CFJr1M2z3SZpQpBMoY3nagpafvS7Uzanq4L7KVXVKi22` | **454** | 3 | **0.7%** | 9.60 | ALICE 30x, METH 17x, ALLINU 7x |
- **`CFJr1M2z` is the exact wallet Ben described: three big wins and completely worthless, because it bought 454 coins to get them.** Keep it in this file as the worked example of why hit rate is the metric.
- **`BXL1iBww` only made $43,000 in total and is still the best wallet found.** Nine coins in its life, three of them 11x or better, 65.5 SOL idle. Size is not the signal, selection is.
- `ECnH8TP` on chain: 833.56 SOL, 105 token accounts, 64 non-empty. `DyCdPoiZ`: 0.57 SOL, 42 accounts. `BXL1iBww`: 65.5 SOL, 34 accounts.

**BOT ROSTER CORRECTED.** `sunicorn` now watches seven: zV18RQ, CAcPyBM3, 2UknaPAY, 9aP2vSFp, ECnH8TP, **BXL1iBww (new)**, **DyCdPoiZ (new)**.
- **REMOVED `Dw3LY8Nn` (104 coins) and `JCLRmm7` (44 coins, 0 SOL custodial, 27% win rate, best win only 6x)** — both added an hour earlier on the multi-win rule and both fail the hit-rate rule. Adding on breadth alone was the mistake.
- Method note: the widening pass that surfaced these was **"any single position at 10x or better on real money"** across the 32 runner coins. It returned 337 wallets, and the long tail is one-hit ANSEM launch snipers at 200x to 800x. **A single enormous multiple is NOT a signal; only two or more on a short coin list is.**

### THE TEN-ROUND SWEEP, 17 Sep 2026 — the definitive unicorn list
- [stated] Ben: "run 10 iterations until you find the very best unicorn wallets. then give me their stats: coins touched, big winners (>10x), last coin touched date."
- **Scale: 10 `market trending` slices (intervals 1h/6h/24h x order-by marketcap/volume/swaps/holder_count x three ATH mcap bands), 88 coins covered, 176 trader pulls, 7,825 unique wallets seen, 63 with two or more 5x wins, 128 fully verified with stats + full book.** Driver at `/tmp/uni/drive.sh`, log `/tmp/uni/drive.log`, report `/tmp/uni/report2.py`.
- **Rounds 1, 2, 4, 5 and 9 added ZERO new coins.** `market trending` at the top of the market is nearly the same set whatever you sort by. **The productive knobs are the INTERVAL (1h added 26 then 19 new coins) and a LOW ATH band, not the sort field.** Do not waste rounds on sort variations.
- **THE SNOWBALL FAILED AND HERE IS WHY: `portfolio holdings` does NOT return `token.address`, only the symbol.** So a winner's coins cannot be fed back as new seeds from holdings. Use `portfolio activity` (which does carry `token.address`) if that expansion is wanted. This blocks the "expand from each unicorn's own winners" idea, which is the way to reach runners that are already dead and no longer trending.
- **`pnl_stat.token_num` IS UNRELIABLE.** One wallet reported `token_num` 3 while its book held 23 positions bought with cash. **Use `max(token_num, positions_bought_from_holdings)` as "coins touched", which is the conservative choice.** Holdings caps at 50 positions, so this understates busy wallets, which is fine because they are being rejected anyway.

**FINAL TABLE. Win = paid at least $300 cash, profit at least $10,000, at least 10x.**
| wallet | coins touched | wins >10x | hit rate | best | made | last NEW coin |
|---|---|---|---|---|---|---|
| `BXL1iBwWwgwfSWTVZpBRfVs5cmvknXNLPeho1AKjzpkY` | 10 | **3** | **30.0%** | 14x | $43,000 | 7 Sep (LEVERCAT) |
| `EK7LgzCoryhTqDysTwzzAdYEQU2Wy4Jj9quk52T1UxkD` | 10 | 2 | 20.0% | 18x | $32,601 | 4 Sep (HYPE) |
| `4Xzmt8Frhf4VMZToogxxxZnY4opWUX9DJyfqupMaxjuH` | 23 | 2 | 8.7% | 47x | $68,758 | 7 Sep (KET) |
| `DyCdPoiZKYyLHwFWgrknKDzKgHd8Zw8UijA59odyzszc` | 28 | 2 | 7.1% | 94x | $147,848 | 17 Sep (STONK) |
| `924nsxPiG31uZv72ga5Lqr9vLkQstF6eh3hFx6VxhN5q` | 28 | 2 | 7.1% | 72x | $68,414 | 12 Sep (Zoe) |
| `ECnH8TPNKQGDdfEE9mkQDFYuMg43kaZz4jjGLaXajcH6` | 28 | 2 | 7.1% | 128x | $515,107 | 13 Sep (ZEC) |
| `8ZXnRVHWzcqgy54q6WUQKfAqfk6fZTYfK9BV37XjNcAf` | 35 | 2 | 5.7% | 28x | $70,682 | 16 Sep (PUMPBALL) |
| `CAcPyBM3t7NhrFDAVxToqoBoM8zumrXf8NPsj7Pn9pa9` | 21 | 1 | 4.8% | 64x | $921,460 | 14 Sep |
- Counter-examples kept deliberately: `ACTbvbNm5qTLuofNRPxFPMtHAAtdH1CtzhCZatYHy831` 1,239 coins for one 25x (0.1%), `Beqv6dz` 800 coins for one 23x, `CFJr1M2z` 454 coins for two.

### ⚠ FUNDING CLUSTERS — some of these wallets are ONE person
**`common.fund_from_address` matched across wallets, so following several of them is following the same operator twice:**
- **`iGdFcQoyR2MwbXMHQskhmNsqddZ6rinsipHc4TNSdwu` funds THREE: `CAcPyBM3t7NhrFDAVxToqoBoM8zumrXf8NPsj7Pn9pa9`, `83kvTTxTJN5UDUTE2qutkVNqRmx2qkabNhPx8smLdsbN`, `GZetT3iRmhuzuT1gkSk69MXMe8hRJZJGz1PPdKW1J2vq`.**
- **`B48kNVXs4YK4amkBCH2XokQiv1SeiVQGHDR17xDeKAAn` funds TWO: `9aP2vSFp2rydWR7C7kVZSQLQa2sT62w23ZHVP8BVSR5T` and `4Xzmt8Frhf4VMZToogxxxZnY4opWUX9DJyfqupMaxjuH`.**
- **RULE: always diff `fund_from_address` across a candidate list before calling them independent confirmations.** Three wallets in the same coin is only a crowd signal if three different people are behind them.
- `sunicorn` now watches TEN wallets: zV18RQ, CAcPyBM3, 2UknaPAY, 9aP2vSFp, ECnH8TP, BXL1iBww, DyCdPoiZ, 924nsxPi, 8ZXnRVHW, EK7Lgz.

### ROUNDS 11-20, 17 Sep 2026 — THE LIST CONVERGED. Doubling the search found nothing better.
- [stated] Ben: "do 10 more iterations."
- Rounds 11-13 were the **proper snowball**, now possible because `portfolio activity` DOES carry `token.address` where `portfolio holdings` does not: pulled the paged activity feed for the top 20 wallets, extracted **184 coins they had traded that the sweep had never scanned**, filtered out majors and stock tokens by symbol, and hunted the top 59. Rounds 14-17 were four fresh `market trending` slices at 5m and 1m intervals. Rounds 18-20 verified every new candidate.
- **Scale doubled: 154 coins covered (from 88), 12,807 unique wallets (from 7,825), 81 candidates with 2+ wins at 5x (from 63), 91 full books pulled.**
- **AND THE TOP SEVEN DID NOT CHANGE.** The only new entry anywhere in the top 20 was `91dY8u9knzsejMgP7cFAohdug6UzhwszUcMcR3bVjE8L` at a 1.9% hit rate, which is far below the leaders. **That is the most valuable output of the second sweep: the answer is stable, so the list is not an artifact of which coins I happened to scan.**
- Coins touched by several unicorns in the snowball seed and worth noting as a shared hunting ground: OPENAI `PreweJYECqtQwBtpxHL171nL2K6umo692gTm7Q3rpgF` (4 of them), KNOTS (3), $WIF `EKpQGSJtjMFqKZ9KQanSqYXRcF8fBopzLHYxdM65zcjm` (3), ANTHROPIC `Pren1FvFX6J3E4kXhJuCiAD5aDmGEb7qJRncwA8Lkhw` (3), SCHH (3).

### WHAT THE UNICORNS HOLD RIGHT NOW (bought with cash, position worth $500+)
- **ZCAT held by FOUR of them**: ECnH8TP $326,616 · DyCdPoiZ $246,265 · BXL1iBww $25,068 · 924nsxPi $17,493. **ZCAT is the single most crowded unicorn position and it is where three of the biggest multiples in the study came from (128x, 94x, 72x, 16x, 14x, 30x).**
- STONK held by three: 2UknaPAY $504,197 · CAcPyBM3 $393,283 · ECnH8TP $163,252.
- ZEC held by three (small), MONEROCHAN by two, mamAOCita by two.
- Single-holder positions: BTC, KET, PENIS, ROCK, KNOTS, MANLET, BULLSHIT, LIT, CATE, GP, XBT, LOOM, Zoe.
- **This "what do several of them hold at once" query is the real product of the whole exercise and should be re-run on a schedule.** Script `/tmp/uni/now.py`.