# #724 v3 thin seeds replay (analyst1, 21 Oct 2026)
Source: v3hist.db (v3 pools with swaps, 4945 pools with a swap at birth+30..90 s). Entry = first swap price in that window. Exit = 5x take-profit, else -50% trail from running peak, else last price; any liquidity `remove` between entry and exit = 0.05x. Code: /opt/crew/work/analyst1/r724.py.
Hunt: enter 30-90 s, target 5-20M tier.

| first-swap cap | n | mean | median | ex-top3 | later-half holdout | pulls | hit 5x |
|---|---|---|---|---|---|---|---|
| <3k (thin) | 425 | 1.33 | 1.00 | 1.30 | 1.41 (n213) | 61 | 91 |
| <1.5k | 119 | 1.57 (no pull model) | | | | | 17 hit 5x, 13 peaked 10x+ |
| 3-10k | 2228 | 2.21 | 1.31 | 2.21 | 2.27 | 136 | 662 |
| >10k | 2285 | 2.13 | 1.49 | 2.13 | 2.22 | 162 | 490 |

Outside-seller check (>=2 distinct non-creator sellers within -2..+10 min of the 5x touch, else tp counted 0.3x): thin 1.47, 3-10k 2.20, >10k 2.07 (before the pull model).

## Findings
- Thin seeds are NOT the best bucket. They rug more (14% pulled vs 6-7%) and mean is lowest (1.33). The 5-20M runners come from the whole v3 population, not from tiny seeds.
- Entering any v3 pool at 30-90 s with 5x tp + 50% trail shows mean ~2.2, stable ex-top3 and in the later-half holdout. This agrees with the verified v3sparse60 family.
- CAVEATS (why this is not yet money): fills at the swap price with no slippage, tp assumed filled at exactly 5x on a pool of a few ETH (real depth will cut it), pools without a swap in the window are skipped (survivor bias small: they were not tradable anyway), pulls detected only inside entry..exit.
## Verdict
Thin-seed idea (#724) = discard as a separate filter. Next step for the 30-90 s all-v3 entry: depth-aware fill (v3depth.db) on the 3k+ buckets; belongs to the sparse60 lane owner, not a new test.