# Copy-graph edges — who copies whom (2026-05-30)

**Phase 3.** Resolves the directed structure the aggregate copy-graph collapsed:
(A) which of our 54 alpha leaders each copier follows, and (B) each copier's
*true* dominant upstream wallet (any wallet, not just our 54). Full addresses.

| Query | id | exec | cost |
|---|---|---|---|
| A — per-edge copier→leader | 7614882 | 01KSWBV6A2WN2PZWQEHFSARHJY | 9.4 |
| B — reverse-upstream | 7614898 | 01KSWC07P4JS6BXMBKTM4F9XBE | 11.8 |

`%feed` = share of that leader's entire 30d token feed this copier buys.
`s` = median slots behind. (C) on-chain.

---

## A. Directed copier → alpha-leader edges (within our 54)

**Tight 1:1(-ish) relationships — genuine directed copytrading:**

| Copier | Primary leader(s) followed | Strength |
|---|---|---|
| `7NAd2EpYGGeFofpyvgehSXhH5vg6Ry6VRMW2Y6jiqCu1` | `FxN3VZ4BosL5urG2yoeQ156JSdmavm9K5fdLxjkPmaMR` **49% feed** + `2fg5QD1eD7rzNNCsvnhmXFm5hqNgwTTG8p7kQ6f3rx6f` **48%** | near-exclusive, s15 |
| `suqh5sHtr8HyJ7q8scBimULPkPpA557prMG47xCHQfK` | same pair: `FxN3VZ4B…` 38% + `2fg5QD1e…` 37% | near-exclusive, s15 |
| `FjNJNHvXA5aqKYG2Qoo4UddFg5oNdZPddNPCU7uQHTBc` | `omegoMAe1AMY5MFKQQr3JwXVy8F4eCvmBAfcpo8XAfq` **17%** | dominant single, s4 |
| `ARu4n5mFdZogZAravu7CcizaojWnS6oqka37gdLT5SZn` (whale) | `4vw54BmAogeRV3vPKWyFet5yf8DTLcREzdSzx4rw9Ud9` **43% feed** (+ 25 others) | heavy, s14 |

**The STorreSu / 1aDerPKk sub-community** (Jito tail-farmers — a distinct
cluster that copies a *different* leader pair than everyone else):

| Copier | Follows | |
|---|---|---|
| `kEFiAX3jo5NmemysQov342TZ9mGh6yp92GDRjhA8XDf` | `STorreSu8X6yPLmiEScNHSGDinqV7j84hWjmvH9SPwk` 31% + `1aDerPKk87xJHCAqTY5bGxF5Xet2MG476y4Zmq2WJ8Y` 30% | s10–11 |
| `6XzmFvaNivoYZATpfjLmU9yWFU8p1XTiDnjFJQ4JHFGd` | `STorreSu…` 28% + `1aDerPKk…` 20% | s11–13 |
| `4G9RznKvrtBWn2sRETcAsnSRkectXiDsFP5imBbSiFM2` | `STorreSu…` 19% + `1aDerPKk…` 17% | s10–13 |
| `BeUW3MFm5Em8zpReaSrfgRNyHhya7jA5VAJdnfEk7HoU` | `STorreSu…` 18% + `1aDerPKk…` 16% | s11 |
| `912iwi9rQV6mc6RxGa77QDjcQjLgoKvEVahBFeqdpSgN` | `STorreSu…` 20% + `1aDerPKk…` 18% | s13–14 |

**Everyone else** = broad fan-out: ~15–17 leaders each, but each edge is only
0.5–8% of the leader's feed → these are *not* dedicated to any one alpha. The
recurring anchor across almost all of them is `64hP97Bwr5PubotcTeGgfhkFrGiLVVxT2kVo9M9b4AEz`
(highest token volume, so highest absolute overlap) at low %feed.

**Same-block (slot-1) copiers** concentrate on the high-volume leaders:
`5veTCy9eDaL66LqABfpywN6jr1s7zaXP4uit8sMRAaHA` → `SQHK48QT8SY1vYN44iXji7wQ6CJek8AjfX6mBp47TZq`
(12% feed, **s1**) + `DbEh3Yah8wPtRNpcxtixHMP8Kw1H2wvfv9D1DLVH2zKH` (s1);
`HrLzB8CiQnWCnLtzKrSHSjU6PisSU4FzAkMUeZbCK4ja` → `64hP97Bw…` (s1);
`AwfBEu6apugComhG3LFY12DQrTyXMY9syuvCgxHLLWzL` → `SQHK48QT…` (s1).

---

## B. Reverse-upstream — each copier's *true* dominant source (any wallet)

For the top-20 copiers by PnL, the wallet(s) that most consistently buy 1–15
slots **before** them. **No source exceeds ~25% of a copier's book** → they are
not single-wallet mirrors.

The recurring upstream cohort (appears as a top source for many copiers):

| Upstream wallet (full) | In our set? | Role |
|---|---|---|
| `haqqiUUXB5gp7gDbSjpEkw3phAjQb25r86uXJsZQgWq` | copier #18 (idx **411**, front-of-slot) | #1 source for ~half the field — fast, everyone trails it |
| `EgQX9R3Qph1dPHE1Ysou1auSYqRGomCNmLDC28Yg77aq` | LAUNCH INSIDER | precedes *everyone* (snipes genesis) — upstream by construction, not copied |
| `97V26zCSXhBRSGsnfK4dTai8sd34UHQkA3G7jS9MG9BC` | copier | fast wallet others trail |
| `nya666pQkP3PzWxi7JngU3rRMHuc7zbLK8c8wxQ4qpT` | copier | fast (s3–7) |
| `Etrk5GiyQSJp2Jgu7A6GJqV7rwSc3PfgcYb2uDe8PTMJ` | not in set | fast upstream wallet |
| `ARu4n5mFdZogZAravu7CcizaojWnS6oqka37gdLT5SZn` | whale | upstream of the slower copiers |
| `BwWK17cbHxwWBKZkUYvzxLcNQ1YVyaFezduWbtm2de6s` | **NEW (not in set)** | ARu4n5mF's #1 source (11.9%) — uninvestigated |
| `BnHZ99JebZEw17DXQnrnZeF3VqEBLigCNxxmv8GcobKC` | **NEW** | recurring source |
| `DQApNebkK5wCmy92T5c4Z9NePWjP7dC4avpLMzuxadxR` | **NEW** | recurring source |
| `Anubis512ho5t7S6LNSwoxUWdeQmX2kf3RvZ8ApHHF5w` | **NEW** | recurring source |
| `3KcPSJ8ouE7H1feoWHmhDwXhLnXhpxP6R6713uytFmBM` / `BS7kabqFK8Qb2AqrzLC1M1VLhrh79QyoyQkJMossVDY9` | **NEW** | kEFiAX3j's top sources (Jito sub-community) |

Notable exception — a tight pair surfaces in B too:
`suqh5sHt…`'s sources include `FxN3VZ4B…` (9.7%) + `2fg5QD1e…` (9.7%) +
`7NAd2EpY…` (9.7%) — confirming the FxN3VZ4B/2fg5QD1e/7NAd2EpY/suqh5sHt
micro-cluster from Query A.

---

## Synthesis — what the copy structure actually is

1. **The field is a diffuse momentum cascade, not 1:1 copytrading.** For most
   copiers no single wallet explains >~15–25% of their book; they pile into a
   token once a *cohort* of fast wallets has bought, 5–9 slots back. "Copy
   wallet X" is the wrong model for ~80% of them.
2. **A small number of genuine 1:1 relationships do exist** — the
   `FxN3VZ4B…`+`2fg5QD1e…` ← `7NAd2EpY…`/`suqh5sHt…` cluster, and the
   `STorreSu…`+`1aDerPKk…` ← Jito sub-community. These are the only places a
   "copy a specific wallet" thesis is on-chain-supported.
3. **The real upstream originators are fast wallets, not our alpha set** —
   `haqqiUUX…` (front-of-slot idx 411), `EgQX9R3Q…` (genesis sniper), `97V26zCS…`,
   `nya666pQ…`, plus uninvestigated newcomers `BwWK17cb…`, `BnHZ99Je…`,
   `DQApNebk…`, `Anubis512…`. **These — not the slow copiers — are where the
   timing edge originates.** Worth a discovery pass of its own.
4. **`EgQX9R3Q…` as universal "source" is an artifact** — a launch sniper is
   early on everything, so it precedes everyone without copying anyone. Don't
   read it as a leader being copied.
5. **"Why" stays off-chain.** We now know *what* each copies (cohort/leader,
   latency, size) but never their internal selection model.

## Open / next
- Discovery + screen on the **upstream originator cohort** (`BwWK17cb…`,
  `BnHZ99Je…`, `DQApNebk…`, `Anubis512…`, `Etrk5Giy…`) — these may be better
  alpha candidates than the slow copiers.
- Per-edge directed-copy confidence (slot-delta variance) to separate true
  mirrors from cascade followers.