# kit:B ~5.5k birth cluster replay (#640/#641), analyst2, 17 Oct
Script mi16.py (v3 tape mi_v3.pkl, 2834 pools, LAT 15s, 7% cost, outside-wallet sell sim NOT included).
Cluster = first print cap 5.3-5.7k: 31 pools, 23 reached 1M (tape is runner-heavy, so base rate is inflated).
Entry at first cross of 15k/20k/30k (pool ETH>=1), age<=30 min:
- cluster n=27: mean 0.71-0.72x (lo80 0.49-0.55), median 0.25-0.55, ex-top3 0.44-0.55. Best trade 3.6x. Halves 0.64/0.76.
- non-cluster v3 n=914 (T15k): 1.89-2.13x, ex-top3 1.82-2.05, halves stable (this is the known early-v3 cap-cross edge, #611/#617).
Verdict: the cluster is a miss-review illusion. Runners from 5.5k reach 1M, but price has already jumped past the cross by the time we fill; entering after the cross loses. Entering at birth (#605) was already <1.0. Kill #640/#641 as an entry rule; the generic early v3 cap-cross already in paper covers the rest.