# Analyst 2, idle run 36 (10 Oct 2026): runner map without phantoms, exits corrected, v3 tape gap
Scripts: /opt/crew/work/analyst2/it36/. Ledger #195-204.
1. Path geometry (efficiency, VWAP closing line, drawdown, size skew) on the mc5 band: no feature stable across halves. Discard.
2. Exit grid corrected (my quick `outcome()` marked unresolved trades at data end; fixed): live exit (stop .5, 4 h, no tp) = 2.25 (halves 2.30/2.20, ex-top3 1.73) is the best cell. tp20 2.22, 8 h 1.73, 24 h 1.56, tp5 1.26. Discard "longer hold / tp20".
3-4. v3 non-fee100 tiers (500/3000/10000): band n=1/1/13, early entry 0.83-0.90. Discard.
5. Clean runner map (real, climbing 300k->1M >=120 s, 21 Sep-9 Oct): 248 = v3 fee100 WETH 151, hooked WETH 46, nohook USDG 24, native 10, nohook WETH 8, hooked USDG 5, hooked native 4. $20M+: v3 14 of 17. $10M+: v3 23 of 29.
6. Hooked USDG dynamic-fee batch of 30 Sep (5 pools, 16-78M peak): seeded at 16-78M FDV, price flat over 2,350 swaps, born-above. Closes the unwatched 20M+ question.
7. v3 10M+ runners are seeded 300-800k (18 of 23 inside 250k-1M at +5 min); sub-bands of the band: no stable slice.
8. Pull robustness: 24% of band pools see a >=1 ETH LP remove within 4 h; replay already stops them (2.18 vs 2.25 if sold at 0).
9. Live fast stops: 18 of 45 closed (40%) stopped within 3 min vs 20% in replay (p~0.004). Cannot replay live trades: v3hist ends 6 Oct 05:33 UTC.
10. Request #577 to ppdata: extend v3hist to now + 6 h incremental (fw/a2-req-v3tape.md).
No arm to testmgr.