# A2 iteration run 15 (10 Oct): ENTRY SPEED on the v4 fee-100 unhooked family
Scripts: /opt/crew/work/analyst2/n1-n10.py (PYTHONPATH=. needed). Data: r6weth full series (1387 u100 pools, 16-21 Sep to 7 Oct) + live v4tape (9-10 Oct).

## Finding (READY as paper test A2-FAST5)
Family = v4, fee 100, hooks 0, WETH/native quote (the "u100" group, the operator ramp bot 0x8876...). The first swap is always the ramp bot (500/500 pools). Price then ramps ~2.2x median by +60 s. Our old rule entered at +60 s (mean 0.95-1.0x = no edge). Entering EARLIER, same exit (tp3, stop 0.5, 4 h, pull = 0), same fees (0.01%) and 3% slippage per side PLUS constant-product buy impact (pos/reserve) and sell impact:
| entry after first swap | n | mean | share >=2.5x |
|---|---|---|---|
| +4 s | 460 | 1.71 | 62% |
| +7 s | 587 | 1.60 | 58% |
| +12 s | 865 | 1.46 | 53% |
| +60 s | 1387 | 0.95 | 33% |
- Same 397 pools at every delay (no sample change): +3 s 1.87, +5 s 1.79, +10 s 1.59, +20 s 1.33, +60 s 1.01. Both time halves agree (1.84/1.90 at +3 s; 1.06/0.96 at +60 s).
- Reserve gate at entry (impact aware): >=1 ETH: +4 s 1.89 n=238; >=2 ETH 1.70 n=80.
- Event-time entry (first moment reserve >= 2x seed, median +21 s) 1.46 vs same pools clock +60 s 0.98: it is the same effect (earlier = better), not a separate trigger.
- Jump trap: 0 of 305 tp hits come from a single swap step >=1.5x (median step 1.05x). Median time to tp 118 s (p25 48 s, p90 492 s): exit latency of a few seconds is fine.
- Exits at +5 s entry: tp1.5 1.26, tp2 1.44, tp3 1.68, tp5 1.74, tp8 1.78 (stop 0.5 vs 0.7 irrelevant). Use tp3 (keep consistent with arms) and tp5 as arm B.
- Hooked families do NOT show it: h100 2.37 (+4 s) vs 2.12 (+60 s), H3000 flat. Pure u100 effect.
- Live forward (v4tape, 93 fee-100 unhooked pools 9-10 Oct, price only, 25 min window): peak >= 3x entry 63% at +5 s vs 33% at +60 s = same as history. With a crude liquidity-drop pull check n=24: +5 s 2.78 vs +60 s 2.38 (too small, no depth gate: only direction).
- Loss streak at +5 s: longest run of <1x trades 10 in 500 (win rate 61%). 3%/trade bankroll sim, 100 trades: median x7, p5 x3.6, P(loss) 0%, p95 max drawdown 22% (in-sample, optimistic).
- Feed lag now: v4tape p50 1.47 s, p90 2.14 s (last 3000 swaps); chain block age floor ~0.9 s. So +5 s entry is feasible for paper now; live execution latency is unknown (Ben's sell/fill test).

## Who loses
The operator: its bot seeds and ramps its own pool and pulls it. We buy inside the ramp at the cheap end. At +60 s the ramp has already paid for itself; late buyers carry the pull risk. Kill if forward ON-TIME (feed lag <= 2.5 s) mean < 1.2x over 40 trades, or if the first swap stops being the ramp bot.

## Other results this run (all discard unless noted)
- Recycling (pull of any u100 pool within 120/600/1800 s before launch): 1.60/1.63/1.67 vs 1.73/1.85/1.77. Discard.
- Non-fleet (outside) sender before entry: 15 pools only (2.04 vs 1.67). Too small.
- Creation-to-first-swap lag, seed R0, swaps before entry, reserve at entry: all within 1.4-2.0 without order (R at entry <1 ETH 1.4, >=1 ETH 1.9: use R>=1 ETH as the preferred gate).