# Brief: A2-FAST5 paper test (testmgr)
Evidence: http://159.223.88.168/hx-c9ois/fw/a2-iter15-fast.md
- Universe: v4tape class A1: fee 100, hooks == 0, native/WETH quote.
- Entry: first swap with chain time >= first-swap + 5 s. Only if emitted lag <= 2.5 s (ON-TIME) and reserve >= 1.0 ETH (arm B: >= 0.4 ETH). Size 0.04 ETH.
- Arm A: sell all at 3x entry price (needs reserve >= 2 x size x multiple), stop 0.5, time stop 4 h, pull (liq <= 50% of peak) = 0.
- Arm B: same, tp5. Control arm Z: same pools entered at +60 s.
- Costs: 0.01% fee + 3% slippage per side + buy impact size/reserve + sell impact size*mult/reserve.
- Sim (r6weth, 16 Sep-7 Oct): +4 s 1.71 n=460, +60 s 0.95 n=1387; same 397 pools +3 s 1.87 vs +60 s 1.01; halves agree. Live v4tape peak>=3x: 63% vs 33%.
- Kill: ON-TIME arm A mean < 1.2x over 40 trades, or first swap is no longer the ramp bot 0x8876789976decbfcbbbe364623c63652db8c0904.
- Not for live: execution latency and sell fill unproven; Ben decision after paper.