# A2 miss review of 10 Oct runners (22): what they were and what catches them (11 Oct)
Scripts: /opt/crew/work/analyst2/it74/b.py c.py d.py (radar.jsonl crossings + pool_paths.jsonl).

## 1. The 22 split by REAL market (depth at first crossing), not by the venue label in misses.csv
- 8 v3 no-hook WETH pools (VORTA 20M, hoodcat 17M, PAWS 6.6M, hoodcat 2.9M, DAEMON 1.5M, SIB, roflmao, PAWS 1M): born cap 2.5-11k, depth 30-67k, 300k crossing at age 0, 1M at 24-135 s after. Only 3 of them were ever scored by aipaper (never scored = not in its universe).
- 7 "v4 no hook WETH" labels (FMRGN, CHNCHL, HBLINU, OTLGR, MGCMP, ADWYR, BRML) are actually a hooked WETH pool (v4:weth:hook, depth 83-290k) crossing 300k at 12-36 min; the nohook / native / usdg lookalike pools of the same token are dust (depth 0-9 USD, born at 13-71M cap): their ATH is a one-print artifact. misses.csv takes the dust pool's venue and ATH. These 7 are the H3000-late family (ledger #483, tm-h3late): "not seen by mc5" is correct and irrelevant; the right bot is the 300k-crossing one, and it is already in paper (0 closes yet).
- 3 USDG/SUSD stock-style (QQQ, AAPL, Paper): depth in billions = stable-quoted artifacts, not tradable money. Ignore.
- 4 rest (OP pons, GIGATENEV, HIPPO native hook, 中国人能飞): HIPPO = native hook born 15k, 1M at 5150 s (depth 67k) fits H3000-late style; others dust.
=> Of 22, about 15 are real and tradable: 8 v3 fast ramps + 7 hooked-WETH slow ramps. The review's "7 not seen / 8 rule said no / 7 technical" should be re-bucketed this way; fix idea for review.py: key the runner to the pool with depth>=20k at first crossing, ignore pools with depth<1k.

## 2. New: v3 fast ramp, ENTER AT THE 1M CROSSING (hunt: enter 1M, target 3-10M)
Data: pool_paths.jsonl 21 Sep-11 Oct, v3 WETH nohook, born cap<150k, reached 1M, depth_usd at 1M >=20k (peak = max_cap, peak-based):
- t1m<=60 s: n=44, P3M 57% P5M 43% P10M 34% P20M 16%. t1m<=300 s n=53 55/40/30/15. t1m<=1800 n=107 52/38/26/15. Dumb baseline (all v3 born-small reaching 1M, any depth): n=218 39/29/18/10. So depth>=20k at 1M is the filter (+13 pts P3M), speed adds little.
- Days: 14 of 16 days have P3M>=25% (cells n 1-6). Radar 7-11 Oct, v3 real depth>=20k: n=36, P5M 28%, P10M 25%.
- Other venues at 1M: v4 weth/native/usdg nohook with real depth: n=0..13, P3M 0% (dust). Hooked WETH at 1M (age>=30 min): P3M 50% P5M 10% n=10 (that is H3000-late territory).
Not known: the path AFTER entry (drawdown before 3M), so peak-based P3M 55% says nothing about the stop. Need fill sim: tick path from 1M cross on 53 pools.
RULE for paper (testmgr): v3 WETH no-hook pool, cap_first<150k, radar "cross th 1M" event fires with depth_usd>=20k. Buy at the event (lag matters: radar lag 1-5 s live, backfill bad), size per the usual 0.1 WETH cap. Arms: A sell all at 3M (3x), B half at 3M half at 10M, C stop at 0.5x entry. Time stop 6 h. Kill: mean < 1.3x over 30 closed trades, or fills fail (v3 sells go through the router).
Who loses: the operator pulls LP ~1 h after the run and the 1M buyers who hold; we exit by 3M. Wrong if: depth at 1M is operator-seeded liquidity that is pulled before our sell (check liq events at fill).