# #290 native-ETH tape (r6nat.db, interim) + #285 sellsim_hist: analyst2 verdict, 10 Oct 2026
Scripts: /opt/crew/work/analyst2/it21/n1-n4.py. Sim: pull-aware (cum liq <=10% of max = 0), 3% slip, fee, exit needs in-range ETH >= 2x position, stop 0.5, 4 h hold, entry = first swap >= first swap + delay. Split 60/40 by time.

## 1. #285 sellsim_hist: accepted
tp3 sells ok 47/47 non-failed, median out/model 0.986. The "sell fails at tp3" risk for A1 native is closed. Entry (+5 s) shows 0.21: only a tp exit works, never a forced exit at +5 s.

## 2. Native fee100 unhooked (A1 class, n=1268 at +5 s)
| rule | mean | zero | train / test |
|---|---|---|---|
| +5 s tp3 pos 0.1 | 1.09 | 62% | 1.07 / 1.13 |
| +5 s tp5 pos 0.1 | 1.17 | 74% | 1.26 / 1.04 |
| +60 s tp3 | 0.96 | 63% | 1.00 / 0.90 |
Kill rule I set (native +5 s mean < 1.3): **KILLED as a blanket rule** (WETH same rule 1.7-1.9). 62% of native pools are pulled before tp3.
Separating feature (kill needs spread < 0.3 in both halves): ETH in range at entry. Terciles first half / second half: low 1.29 / 1.20, mid (0.02-0.6 ETH) 1.49 / 1.53, high (>0.6 ETH) 0.35 / 0.70. Initial liq at the 1e23 preset: 0.22 / 0.40 (n 139 / 36), others 1.2-1.3. So spread passes (>0.9, >0.5), but the best bucket is still only ~1.5: filter "skip ETH in range >0.6 at +5 s" removes the worst third; not an edge on its own. Sent as a cheap avoid-arm.
Native fee100 hooked: 0.05 (87% zero): never trade.

## 3. NEW: native H3000 (fee 3000 + hook) is the same edge as the WETH one
Live v4tape: H3000 pools are 87 native vs 6 WETH, so the live paper test sees mostly native. History (r6nat, any hook!=0, n=282): +5 s tp5 pos 0.1: **3.05x**, zero 22%, reach 57%, train 2.42 / test 4.00. Unique-hook pools only (hook used by <5 pools, n=246): tp3 2.25, tp5 3.29 (train 2.79 / test 4.03). Shared-hook pools (n=36, hooks 0x5370.. 29 pools, 0xf752.. 24): 1.3-1.45 = dead, as found on WETH. Forward v4tape native unique-hook (n=10): tp3 1.75, tp5 2.75, tp20 4.35 at +5 s; at +60 s tp5 falls to 1.38 (history: tp5 +60 s 0.84 mixed), so entry speed matters here too.
Unhooked fee3000 native (n=120): 1.5-1.9, test half 2.1-2.6, 31% zero: weaker, watch only.

## What goes to testmgr
- H3000 arms are valid on native pools too (history now n=246 unique-hook, not only WETH n=110). Confirms the existing test; entry delay <= 5-10 s and unique-hook filter stay.
- Avoid-arm for A1 native: skip if ETH in range at +5 s > 0.6 (or initial liq preset 1e23).
Nothing here is for live: Ben's sell test for H3000 stays the gate (outsiders' sells, not covered by sellsim_hist, which is A1 only).
Caveat: tape ends 9 Oct 21:30 UTC (interim); rerun on the final refresh.