# A2 idle run 20 (10 Oct): sellsim_hist verdict, native-quote A1 class, 8 discards
Scripts: it20/s1.py t1.py t2.py f1-f3.py (research-box /opt/crew/work/analyst2).

## Findings
1. SELLSIM_HIST (my request, now delivered): tp3 sells 44/45 ok, median out/model 0.986 (p10 0.99). The "sell at tp3 is real" condition of FAST5 passes. Live sellsim on 20000 s old pools fails (pools pulled), irrelevant for tp exits.
2. NEW FACT: live fee-100 hookless v4 pools (A1) are 77% NATIVE-ETH quoted (142 of 184 today); history/tests (r6weth) cover WETH only. Forward, no survivor filter (v4tape, 25 min hold, +5 s entry, tp3, pull from liq.jsonl):
   - WETH A1 n=40: mean 1.90 (tp5 2.21), zero 4. Same as history (1.8-2.2): FAST5 forward-confirmed on small n.
   - NATIVE A1 n=123: mean 1.49 (tp5 1.67), median 1.01, 36 zeros (29%). Almost all one LP template (sender 0x58daec31, tick_upper 184216): native 1.38, 32 of 61 zero.
   - Pull latency from init: native p10/25/50/75/90 = 9/14/51/281/457 s, 50% pulled within the 28 min window; WETH 13/65/954/1090/1281 s, 19% pulled. Native is a fast-rug mode of the same operator; the +5 s entry sits inside the pull window of 25% of them.
   - Caveat: n small, no reserve/impact in v4tape, 5 s fills assumed.
3. Discards (u100 WETH n=500, y5 = +5 s entry, tp5): token address vanity (flat), hour/weekday (1.4-2.2, no pattern), US cash session 1.68 vs 1.84 (halves disagree), previous-pool outcome / gap to previous launch (1.64-1.88, noise), gap to next launch (look-ahead 1.38 -> 2.21, not tradable; exit-at-next-launch already killed), ramp convexity and reserve at 5 s (flat), stall exit (sell if < 1.0-1.5x after 10-45 s): never beats hold (1.36-1.86 vs 1.77).
## Verdict
No new ready arm. FAST5 stays WETH-only: tell testmgr to NOT run it on native pools until the native tape exists. Data request below.