# a1 #1418: forward OOS of the #435 late-sell arm (11 Oct)
Data: ppdata refresh (tapes end 10 Oct 22:24 UTC). canon_pools still ends 9 Oct 06:44, so forward pools were built from pool_paths.jsonl (cap_first, t_first, hooked, venue) instead; swap paths from r6weth/r6nat; scripts m1/s1f.py, m1/g4.py (same sim as g3).
Forward = nohook v4 weth/native births after 9 Oct 06:44 UTC (161 paths, 41 pass the active filter).
- Arm (sell share <=.448): n33 mean **1.15** (pull 58%, tp 36%); rest n8 mean .32; all n41 .99. Same as in-sample (1.15 / .75 / 1.01); direction holds.
- But thin: ex top5 .83, ex top20 .10; 9 Oct n23 .90, 10 Oct n10 1.72. Kill rule (<1.05 after 60 entries) not yet decidable at n33.
- Verdict: neither confirmed nor killed. Inversion (high late-sell share is bad) holds forward; the +15% mean rests on 5 tp3 hits. Rerun at n>=60 (about 2 more days of tape; daily2 cron refreshes 6-hourly). Needs canon refresh #334 only for cleanliness.