# A1.11 Self-buy check on window 2 (Analyst 1, 6 Oct 2026, 14:00 Bangkok)

## Read this first
- **CORRECTION (A1.13, 15:30 Bangkok): the 6 "farm wallet" addresses below are the coins themselves (token contracts, 3,554 bytes), not wallets.** I printed coin addresses where wallet addresses belonged. Rerun with every contract removed: the filter result is unchanged (49 coins at 2.08x, A 2.18x, B 2.01x). The real overlap is hundreds of wallets each in 2 of the flagged coins, many at a near-equal 1.4 to 1.5 ETH. See http://159.223.88.168/hx-c9ois/sw/a1-13-money-map.md
- **The v3 rule is mostly clean, but 2 small wallet farms got through. A new entry filter catches them and lifts the result in both halves.**
  - In 51 of the 57 coins (gate data for the 58th is missing), repeat wallets bought under 10% of the pre-gate buy ETH. The median is 0.1%. So "sellable $1M" is real money in most coins.
  - **6 coins (all 21 to 24 Sep, half A) were pumped by 2 farms.** Each farm used a fresh set of wallets that bought in exactly 3 coins:
    - Farm 1: `0xcefbfccd83443e7aa1c493f4df0036013542f301`, `0xcc4cb99ef15e9799e1d7e45058683a72daf5b022`, `0xabe99cf268cdc8bd70a71a5283fdacbcb36714fd`.
    - Farm 2: `0x55683371036401f8cae676c260efcde9be36a7f4`, `0xc3ba8a6f8d011881ceca8ad6e2b524232a40aa26`, `0x5cee66125e28aec539365c7a34fd25a2210c3966`.
    - In those 6 coins, repeat wallets bought 32% to 95% of the pre-gate ETH. 5 of the 6 trades were stops (0.56x), and 1 hit target (4.82x).
  - **The army filter (0.009) misses farms like this.** It only counts wallets seen in 3 or more EARLIER coins. A farm wallet's 3rd coin has only 2 earlier coins behind it, so it never trips. In walk-forward terms the 6 farm coins scored 0.000 to 0.004 on army.
- **New entry filter, walk-forward:**
  - Skip the coin if more than 5% of its pre-gate buy ETH comes from wallets that already bought pre-gate in at least ONE earlier gated coin.
  - It drops 8 coins that averaged 1.08x.
  - Kept: **2.08x** (half A 2.18x n20, half B 2.01x n29), vs 1.94x for all 57 (A 1.91x, B 1.96x).
  - Cuts of 5% and 10% give the same answer. 20% keeps one more coin (2.05x).
  - The real $20M coins in the rule set (9 of Analyst 2's 24) all score 0.0% to 0.5%. The filter doesn't touch the winners.
- **Hold side (virality "new buyers per hour"): repeat wallets can fake it.**
  - In 12 coins, more than 20% of the "new" buyers in hours 0 to 4 after the gate were wallets already seen in an earlier gated coin. In `0xc3ba8a6f8d011881ceca8ad6e2b524232a40aa26` it was 95%.
  - Those 12 averaged 1.17x, vs 2.14x for the rest (both halves: A 2.04x, B 2.22x).
  - This data comes after our entry, so it can't be an entry rule. But it should be a hold or sell check: "new buyers" only count if they are new to the chain's recent gated coins.

| Filter (walk-forward) | Coins kept | Kept result | Half A | Half B | Dropped coins | Dropped result |
|---|---|---|---|---|---|---|
| none (A2.8 strict) | 57 | 1.94x | 1.91x | 1.96x | | |
| army 0.009 (already in the rule) | 57 | 1.94x | 1.91x | 1.96x | 0 | |
| pre-gate repeat (1+ earlier coin) over 5% | 49 | **2.08x** | 2.18x | 2.01x | 8 | 1.08x |
| same, 2+ earlier coins over 5% | 55 | 1.99x | 2.02x | 1.96x | 2 | 0.57x |
| hold check: new buyers h0 to 4 repeat over 20% (after entry, test only) | 45 | 2.14x | 2.04x | 2.22x | 12 | 1.17x |

## Method
- **Buyers:** every token Transfer for the 57 rule coins, from the pool's first block to gate + 4 h (541,381 transfers, read from chain). A buy is a transfer out of the pool. When the tokens pass through a router in the same transaction, I follow them up to 3 hops to the final wallet (1.2% of buys).
  - 23 contracts among the 300 biggest receivers (routers, aggregators) are excluded.
  - Buy ETH comes from the same transaction's swap.
- **Pre-gate** = pool's first block to the gate block. **New buyers h0 to 4** = a wallet's first buy in this coin falls 0 to 4 h after the gate.
- **Repeat, walk-forward:** a wallet counts only if its earlier buy in another gated coin came before this coin's gate block. Memory = the rule's own gated coins only, which a live bot knows in real time.
- **Hindsight check:** 685 wallets bought pre-gate in 3 or more of the 57 coins, and 25 wallets in 10 or more. The busiest is `0x1364def0021d5137f30b315adf58674a8b8d0f7b` (24 coins). These big repeaters are tiny buyers (under 0.1% of pre-gate ETH), so they don't move the gate. Only the two farms do.

## Limits
- 8 dropped coins, mostly in half A. Half B gains only 0.05x, so the filter's value rests on catching farms like the 21 to 24 Sep ones again. It should run as a tagged paper variant first.
- The Virality Expert's 569-coin set is not public, so the $20M check covers only the 9 v3 rule coins.

## Asks for Supervisor
- Coder: add `rep1_pre` (share of pre-gate buy ETH from wallets seen pre-gate in 1+ earlier gated coin) to every v3 candidate. Paper variant: skip if over 5%.
- Virality Expert: recompute "new buyers per hour" with wallets seen in earlier gated or $500k coins removed, and report how many ladder passes change.

First principles score: 85%. The buyer list comes from on-chain transfers followed to the final wallet, and the filter uses only data before the gate. Missing: the farms' funding sources, and a larger sample (2 farms, 6 coins).

Scripts: /root/a1/a111/fetch.py, build.py, an.py (review container). Latest: http://159.223.88.168/hx-c9ois/sw/a1-11-self-buy-v3.md