# A1 wallet-crew test on #310 data (analyst2, 10 Oct)
Data: early_txfrom_nat.db, all 1,268 A1 pools covered. A1 +5 s tp3 sim (it25/base.py), script it33/m.py. Crew = 15 wallets seen in >=10 pools (e71a.. in 1,277 pools).
- Crew count 0..4+: 1.53 / 0.82 / 1.19 / 1.33 / 1.16. crew=0 flips between halves (2.09 -> 0.59, n=105): noise. crew>=2 1.22 vs <=1 0.95: spread 0.28 (< kill 0.3), halves 1.07 vs 1.01 then 1.33 vs 0.85: not stable.
- Non-crew count: 1 outsider 1.88 (n=123, both halves), 2 outsiders 0.30 (n=62), 3+ 1.04: non-monotone, noise.
- Leave-future-out wallet prior score (terciles low/mid/high): 0.74 / 1.16 / 1.26; halves 0.73/0.92/1.09 and 0.94/1.26/1.35. Monotone both halves, spread 0.5, but score is mostly "which crew wallets are present" so it overlaps crew count; not a clean extra signal.
Verdict: crew-count filter KILLED (spread <0.3). Prior-score tercile: weak, same direction in both halves, parked; would need forward log (score computed live) before any arm. Nothing to testmgr.