# A1.16 Self-buy / farm check on mc5 (FW Analyst 1, 6 Oct 2026, 15:09 Bangkok)

**Verdict: yes, 4 of the top 8 winners were bought mostly by repeat wallets before +5 min, one of them a known A1.11 farm coin. They still paid at 2 h with real ETH in the pool. A repeat-wallet filter does NOT improve mc5 in both halves, so no filter.**

Definitions:
- **rep1_pre** (A1.11, walk forward): the share of buy ETH before +5 min that came from wallets which bought in the first 15 min of 1+ earlier v3 pool (only pools whose 15 min window had already closed).
- **rep3**: the same, but for wallets seen in 3+ earlier pools (serial snipers).
- Rule tested: mc5, $100, stop 0.5x, hold 2 h. Fixed pull rule from A1.14.

## Top 8 winners
| pool | token | x | buyers before +5 min | fresh | rep1 | rep3 | buy ETH | day |
|---|---|---|---|---|---|---|---|---|
| 0xb4b0dfeaa2e2a838f0f7a304f005ebe32390d4ef | 0xc38c332012a9116dcade0c7f43b5d3c71799a9f8 | 35.6 | 57 | 47 | 0.00 | 0.00 | 94.6 | 6 Oct |
| 0xeb1646dbfbeacf76483df1092c208388c029c534 | 0x257723230c9cccf6a2df1e7cf67cffcfcf01eeb8 | 21.1 | 96 | 1 | **0.77** | 0.00 | 61.7 | 30 Sep |
| 0x6e8a6a1af8ec53f3f09c7303e5403558dee42d0a | 0xba1a2d9783ebe2b76493c5c12eb9813a0e062843 | 20.3 | 34 | 28 | 0.01 | 0.01 | 65.6 | 2 Oct |
| 0x7608c38ea8cc2082b043e886185979942cdeb58e | 0x8e41fd25e85e624560fa50ff9f117a02cd30b8b3 | 16.0 | 267 | 264 | 0.00 | 0.00 | 55.2 | 4 Oct |
| 0xb6fe1035f6edfaeb7a9f7d4c19e11d35b3bd7c57 | 0xabe99cf268cdc8bd70a71a5283fdacbcb36714fd | 14.9 | 68 | 52 | **0.63** | 0.00 | 17.7 | 23 Sep |
| 0xc586fcefc588870aa3159b713aa813578326562c | 0x88042d98421b4d4014910bd9ca40d3cf3a309733 | 9.5 | 176 | 153 | 0.08 | 0.00 | 29.1 | 22 Sep |
| 0xd958ef2cc262fdbe3b670aec434d8b9102f5ad34 | 0xfa1824e750f9a5bc7d253342935471b8f2709d70 | 8.8 | 160 | 3 | **1.00** | **0.71** | 126.7 | 23 Sep |
| 0x1a2822ccba364190fe945e80c14493c3964a494d | 0xe872a965b354a86545520360011fb824ed27db6c | 8.7 | 83 | 6 | **0.81** | **0.81** | 39.0 | 25 Sep |

- 0xabe99cf268cdc8bd70a71a5283fdacbcb36714fd (NOSH) is one of the A1.11 farm 1 coins.
- 0x257723230c9cccf6a2df1e7cf67cffcfcf01eeb8 had 95 of 96 early buyers seen in earlier coins, but none in 3+ coins. That looks like a fresh farm cohort reused once.
- 0xfa1824e750f9a5bc7d253342935471b8f2709d70 and 0xe872a965b354a86545520360011fb824ed27db6c were bought mostly by serial snipers (rep3 0.71 and 0.81).
- All four were still sellable at 2 h: the pool held 34 to 68 ETH and had zero Burns after entry (A1.14). The farm built a real run and the outside money came in.

## Does a filter help? (mean / without top 3; H1 n, H2 n)
| set | H1 | H2 |
|---|---|---|
| all | 1.86x / 1.15x (42) | 1.58x / 1.12x (159) |
| skip rep1 > 5% | 1.21x / 0.35x (14) | 2.22x / 1.04x (58) |
| skip rep1 > 20% | 1.54x / 0.54x (19) | 1.96x / 0.99x (71) |
| skip rep1 > 50% | 1.43x / 0.61x (23) | 1.66x / 0.95x (97) |
| keep only rep1 > 20% | 2.12x / 0.93x (23) | 1.28x / 0.87x (88) |
| rep3 under 5% | 2.24x / 1.04x (23) | 1.77x / 1.07x (105) |
| rep3 20%+ | 1.50x / 0.45x (14) | 0.90x / 0.59x (41) |

- rep1 flips sign between the halves: a filter would lose in H1 and win in H2. It is not stable, so drop it.
- rep3 under 5% (no heavy sniper share) is better on the mean in both halves. Without the top 3 it is the same as no filter (1.04x / 1.07x vs 1.15x / 1.12x), and it halves the trade count. It does not justify losing the trades. Revisit once 30+ live trades exist.
- **Keep the rule unfiltered.**

Code: a1code/v3self.py (research-box /root/fwa1).