---
name: copytrade-targets-3
description: Continuation of copytrade-targets-2.md from 17 Sep 2026. The funder-grouping idea, the RETRACTION that killed it (fund_from_address is almost always an exchange, not a person), the exchange-vs-person test that replaced it, and Ben's locked keep-list.
sources: [cowork]
aliases: [copytrade part 3, unicorn funders, funder scoring, keep list, exchange funder test]
---
Continues [[copytrade-targets-2]]. The unicorn definition (v4) and the ten-round and twenty-round sweeps are in that file.

# 17 Sep 2026. FUNDER GROUPING, AND THE RETRACTION THAT FOLLOWED WITHIN THE HOUR

- [stated] Ben: "You can keep the multiple wallets if it's the same funder, but what we care about is the total coins touched to win ratio across all of the funders wallets, this applies to other funders with multiple wallets also."
- The idea is right: a person who runs five wallets and shows you the good one is an editor, not a picker. **The implementation was wrong, and the wrong version reached Ben.**

## ⛔ RETRACTED IN FULL: "the Original Unicorn's operator is a 0.7% lottery machine"
**I told Ben that `iGdFcQoyR2MwbXMHQskhmNsqddZ6rinsipHc4TNSdwu` runs five wallets totalling 687 coins for five wins, and that `zV18RQ` and `CAcPyBM3` are therefore the same person. ALL OF THAT IS FALSE.**
- **`iGdFcQoy...` holds 66,661 SOL and does 4,802 transactions per day. It is an EXCHANGE HOT WALLET.** The five wallets sharing it are five strangers who happened to withdraw from the same venue.
- Same for every other multi-wallet "funder" in the data. Measured balance and transactions per day:
  | funder | wallets | SOL | tx/day | what it really is |
  |---|---|---|---|---|
  | `5tzFkiKscXHK5ZXCGbXZxdw7gTjjD1mBwuoFbhUvuAi9` | 8 | **2,006,551** | 25,899 | exchange |
  | `iGdFcQoyR2MwbXMHQskhmNsqddZ6rinsipHc4TNSdwu` | 5 | 66,661 | 4,802 | exchange |
  | `AxiomRXZAq1Jgjj9pHmNqVP7Lhu67wLXZJZbaK87TTSk` | 4 | 5,150 | 36,348 | **Axiom, the trading terminal** |
  | `F7p3dFrjRTbtRp8FRF6qHLomXbKRBzpvBLjtQcfcgmNe` | 3 | 5,023 | 318,819 | service |
  | `MfDuWeqSHEqTFVYZ7LoexgAK9dxk7cy4DFJWjWMGVWa` | 3 | 91 | 200,000 | service |
  | `4NyK1AdJBNbgaJ9EsKz3J4rfeHsuYdjkTPg3JaNdLeFw` | 2 | 20,803 | 7,912 | exchange |
  | `B48kNVXs4YK4amkBCH2XokQiv1SeiVQGHDR17xDeKAAn` | 2 | 1,131 | 4,456 | exchange |
- **NOT ONE multi-wallet funder in the entire dataset is a person. Every wallet found is an independent trader.**
- **Therefore: `zV18RQ` is NOT part of a 687-coin operation. Keeper 3 does NOT fall to 5.3%, it stays at 20%. `zV18RQ` and `CAcPyBM3` are NOT the same person.**

## THE TEST THAT REPLACES IT — run this before EVER grouping by funder
Solana RPC, free, container-side: `getBalance` and `getSignaturesForAddress(limit=1000)`, then transactions per day from the first and last `blockTime`.
- **tx/day > 200, or balance > 500 SOL -> EXCHANGE OR SERVICE. The funder is not an identity and must never be used to group wallets.**
- tx/day 20 to 200 -> probably a service, treat with suspicion.
- **tx/day < 20 and a small balance -> a real person.** Every single-wallet funder in this dataset measured 0.3 to 4.7 tx/day, i.e. all real people.
- Script `/tmp/fclass.py` in the Cowork container. **This is now a mandatory gate on any wallet-clustering claim, alongside the sweep-fleet test in [[copytrade-targets-2]].**
- **The wider lesson, and it is the third time this pattern has bitten: a shared address is not a shared identity until its ACTIVITY RATE says so.** The same mistake shape as reading a router as one entity splitting bags, and as reading 1,378 airdrop legs as accumulation.

## ⚠ A THRESHOLD I MOVED WITHOUT SAYING SO, and Ben caught it
- [stated] Ben: "And before, you told me that this Original Unicorn wallet had multiple big wins. Why does it now say just one?"
- **He was right.** I called it two wins at a "3x and $50k" bar (STONK 75.7x and ALLINU 6.0x). When he asked for winners **over 10x**, ALLINU at 6.0x stopped counting. Same wallet, same trades, different bar, no warning from me.
- **RULE: whenever a threshold moves, restate the counts under BOTH bars in the same sentence.** Quote wins at 3x, 5x and 10x side by side.

## THE KEEP LIST Ben locked (never discard, keep appending). CORRECTED NUMBERS.
| label | wallet | coins touched | wins >10x | wins >5x | hit rate (10x) | the wins |
|---|---|---|---|---|---|---|
| **ORIGINAL UNICORN WALLET** | `zV18RQeme9fDZUbCGf7skSWNz6Vphs5j3PQUHjHdcgi` | 24 | 1 | 2 | 4.2% | STONK 75.7x, ALLINU 6.0x |
| **KEEPER 2** | `BXL1iBwWwgwfSWTVZpBRfVs5cmvknXNLPeho1AKjzpkY` | **10** | **3** | 3 | **30.0%** | ZCAT 14x, LEVERCAT 12x, AGI 11x |
| **KEEPER 3** | `EK7LgzCoryhTqDysTwzzAdYEQU2Wy4Jj9quk52T1UxkD` | **10** | **2** | 2 | **20.0%** | UBI 18x, PURR 14x |

## LEADERBOARD, wallet-level (the correct unit), 17 Sep 2026
| wallet | coins | 10x+ | hit | the wins |
|---|---|---|---|---|
| `BXL1iBwWwgwfSWTVZpBRfVs5cmvknXNLPeho1AKjzpkY` KEEPER 2 | 10 | 3 | 30.0% | ZCAT 14x, LEVERCAT 12x, AGI 11x |
| `EK7LgzCoryhTqDysTwzzAdYEQU2Wy4Jj9quk52T1UxkD` KEEPER 3 | 10 | 2 | 20.0% | UBI 18x, PURR 14x |
| `4Xzmt8Frhf4VMZToogxxxZnY4opWUX9DJyfqupMaxjuH` | 23 | 2 | 8.7% | KET 47x, UBI 13x |
| `DyCdPoiZKYyLHwFWgrknKDzKgHd8Zw8UijA59odyzszc` | 28 | 2 | 7.1% | ZCAT 94x, PENIS 15x |
| `924nsxPiG31uZv72ga5Lqr9vLkQstF6eh3hFx6VxhN5q` | 28 | 2 | 7.1% | GP 72x, ZCAT 16x |
| `ECnH8TPNKQGDdfEE9mkQDFYuMg43kaZz4jjGLaXajcH6` | 28 | 2 | 7.1% | ZCAT 128x, STONK 28x |
| `8ZXnRVHWzcqgy54q6WUQKfAqfk6fZTYfK9BV37XjNcAf` | 35 | 2 | 5.7% | Chonketha 28x, FRANKIE 17x |
| `7y4t9NKPcg6dN3uBgyaDmDHsfedEiNhaCEZZ2y7j5Dc5` | 55 | 2 | 3.6% | MENSA 28x, pill 24x |
| `9aP2vSFp2rydWR7C7kVZSQLQa2sT62w23ZHVP8BVSR5T` | 35 | 1 | 2.9% | ANSEM 54x |
| `2UknaPAYpAEP6mbxHRy9mcBgZCHPLWVijQsPBFGJ7Fea` | 35 | 1 | 2.9% | STONK 26x (2,071 SOL idle) |

# 17 Sep 2026, evening Bangkok. THE 20-ROUND SWEEP, AND THE 50-POSITION BUG IT EXPOSED

- [stated] Ben, after the retraction: "Okay, keep running the iterations and keep giving me these clean breakdowns for all the wallets that you find. And then sort them by best performers once you're done."
- [stated] He also fixed the reporting unit: give **above 2x, above 3x and above 10x side by side, every time**, for every wallet. Never one bar alone. (This is the same rule as the threshold-move rule above, now standing.)

## WHAT THE FIRST 20-ROUND ATTEMPT GOT WRONG (iter20.sh)
- Rounds 3 to 20 did **nothing**. The loop was circular: new coins come from following candidate wallets, candidates come from new coins, and both sets were already saturated at 97 wallets / 360 coins. It burned 11 minutes and reported 20 completed rounds that were empty.
- **Fix, and it is the general lesson: a discovery loop needs a WIDER feeder set than its scoring set.** `cands2.py` follows any wallet with one 5x on $5k, or two 3x on $2k. That is 2,255 wallets instead of 97. Scoring still uses the strict bar. After that, every round did real work.
- Also confirmed dead: **`gmgn-cli market trending` is a small live window, not a history.** 36 sweeps across 5 intervals x 5 ATH bands x 6 age bands returned **20 distinct coins, all already scanned.** There are no undiscovered 10M+ runners in that endpoint. The wallet snowball is the only expansion path.
- New tool: `/tmp/uni/athgate.py` gates every discovered mint on **ath_price x total_supply >= $10,000,000** (verified against Fartcoin = $2.6B). Caches to `/tmp/uni/ath.json`.

## ⛔ BUG FOUND AND FIXED MID-RUN: the 50-position ceiling
- `gmgn-cli portfolio holdings` returns **at most 50 positions**. Scoring hit rate off that book silently flatters any wallet that touched more than 50 coins: it only ever sees that wallet's best 50.
- It put three fakes at the top of the board: `CFJr1M2z3SZpQpBMoY3nagpafvS7Uzanq4L7KVXVKi22` (looked 30% above 3x, actually touched 454 coins), `DvyVRgJhnh4f9XfKFMEi8uutHujwLjvQR84ecPWwfxjj`, `Dw3LY8NnG3BB1EEXCNJwAKtVoEE7faEYiygYNHfFPNb1`.
- **Fix: denominator = max(funded positions in book, `pnl_stat.token_num`), and any wallet whose book came back with 50 rows is marked `*` = rate is a CEILING, not a measurement.** `/tmp/uni/report3.py`.
- Pattern, fourth time: **a paginated or capped API turns "what I can see" into "what exists" unless the cap is checked explicitly.**

## FINAL RESULT, 20 rounds, 457 coins, 999 trade histories, 253 wallets scored
Sorted by hit rate above 10x. `*` = book capped at 50, rate is a ceiling. Asleep = no new coin in over a month.
| wallet | coins | above 2x | above 3x | above 10x | best | last new coin |
|---|---|---|---|---|---|---|
| `BXL1iBwWwgwfSWTVZpBRfVs5cmvknXNLPeho1AKjzpkY` KEEPER 2 | 10 | 40.0% | 30.0% | **30.0%** | 15x | 7 Sep |
| `EK7LgzCoryhTqDysTwzzAdYEQU2Wy4Jj9quk52T1UxkD` KEEPER 3 | 10 | 20.0% | 20.0% | **20.0%** | 19x | 4 Sep * |
| `4r9foTryAQaEzUuPWbALHowoaTHN3G7rPkEdvRUvpEF8` | 18 | 11.1% | 11.1% | 11.1% | 11x | 30 Jun ASLEEP |
| `ECnH8TPNKQGDdfEE9mkQDFYuMg43kaZz4jjGLaXajcH6` | 28 | 10.7% | 10.7% | 10.7% | 129x | 13 Sep * |
| `8VknbL73acufE8wBre249QZ2vPZwnxc5ae8Q2dnzHvKr` NEW | 10 | 30.0% | 20.0% | 10.0% | 36x | 14 Sep * |
| `CRCLhwcP1HiWb4p7BL72rBk1WY8HFYPPVeXt7HApu8Qc` | 11 | 9.1% | 9.1% | 9.1% | 80x | 24 Aug * |
| `4Xzmt8Frhf4VMZToogxxxZnY4opWUX9DJyfqupMaxjuH` | 23 | 8.7% | 8.7% | 8.7% | 48x | 7 Sep |
| `924nsxPiG31uZv72ga5Lqr9vLkQstF6eh3hFx6VxhN5q` | 28 | 17.9% | 10.7% | 7.1% | 73x | 12 Sep * |
| `DyCdPoiZKYyLHwFWgrknKDzKgHd8Zw8UijA59odyzszc` | 28 | 7.1% | 7.1% | 7.1% | 95x | 17 Sep * |
| `ACjtaBVhbHdt2tBuYbRWk6WsFrfqmxoLEiRShYYkCiV7` | 33 | 12.1% | 6.1% | 6.1% | 32x | 15 Feb ASLEEP |
| `BrpuFCVpM8kiDTd2Vg2UPW6h3eGhfQNzfG7hARbbneMy` NEW | 17 | 11.8% | 11.8% | 5.9% | 14x | 14 Sep * |
| `8ZXnRVHWzcqgy54q6WUQKfAqfk6fZTYfK9BV37XjNcAf` | 35 | 17.1% | 14.3% | 5.7% | 29x | 16 Sep * |
| `zV18RQeme9fDZUbCGf7skSWNz6Vphs5j3PQUHjHdcgi` ORIGINAL UNICORN | 24 | 12.5% | 8.3% | 4.2% | 77x | 17 Sep * |
| `CR8C7EkkYrxCSfAWV43pq82iMFyc7J794pyGN66qAjgb` NEW | 28 | 14.3% | 14.3% | 3.6% | 37x | 16 Sep * |

- **The single biggest trade in the entire study: `8VknbL73acufE8wBre249QZ2vPZwnxc5ae8Q2dnzHvKr` put $58,860 into ALON and made $2,044,043. 35.7x on a real size, not a lottery ticket.** Only 10 coins touched.
- **The top two did not move across 88, 154, 360 and 457 coins scanned.** Four independent expansions of the coin universe, same answer. That stability is the strongest evidence the list is real rather than fitted.
- The Original Unicorn on its own (its funder is the exchange, so it is a lone wallet): 18 coins with real money in, **3 above 2x (17%), 2 above 3x (11%), 1 above 10x (6%)**. The three: STONK 77x, ALLINU 7x, MONEROCHAN 2.1x.

## SUNICORN BOT (Solana Unicorn Finder) now watches 14 wallets
Added this round: UNICORN 11 `8VknbL73acufE8wBre249QZ2vPZwnxc5ae8Q2dnzHvKr`, UNICORN 12 `4Xzmt8Frhf4VMZToogxxxZnY4opWUX9DJyfqupMaxjuH`, UNICORN 13 `CR8C7EkkYrxCSfAWV43pq82iMFyc7J794pyGN66qAjgb`, UNICORN 14 `BrpuFCVpM8kiDTd2Vg2UPW6h3eGhfQNzfG7hARbbneMy`. Asleep wallets deliberately NOT added.

## PIPELINE, all on Ben's Mac under /tmp/uni/
`cands.py` strict scoring bar -> `cands2.py` wide discovery net -> `act3.sh` paged trade histories -> `runners2.py` unscanned coins good traders paid for -> `athgate.py` the $10M ATH gate -> `token traders` -> `pull2.sh` stats and books -> `report3.py` the 2x/3x/10x breakdown. Driver: `iter20b.sh`. Log: `iter20b.log`.