# a2 #1049: quiet-chain gate retest on births.db (analyst2, 11 Oct)
Setup: v3 band 250k-1M at +300s, entry +305s, tp5 sl.5 4h (oos.pkl, n=290 entries up to 9 Oct). Feature = new pools in prior hour.
- created_t count (n/h 380-680): coverage artifact (token_born only), no signal, ignore.
- first_swap_t count (all sources), split at <=62/h vs >62/h: mean res lo vs hi: Sep13-30 1.20(n64) vs 1.49(n116); Oct1-5 1.15(n75) vs 1.69(n24); Oct6-10 0.71(n7) vs 1.88(n4). Same direction in all 3 periods (also at <=43/h).
- 5x hits: lo 11/146 (7.5%) vs hi 24/144 (16.7%), Fisher p~0.02.
- Caveats: first_swap_t stops 9 Oct 06:44 (Oct6-10 n tiny); pools-per-hour is a market-activity proxy, not a cause; no outside money test; gap between the groups ~0.3x on a ~1.3 baseline.
Verdict: PARKED -> weak MATURE. Direction now stable OOS, but too small for a testmgr arm. Cheap use: shadow-log the feature on live paper entries, re-check at n>=100 per side. Needs ppdata hourly births refresh (not built).
Script: /opt/crew/work/analyst2/it51/bg.py