# A1-ROUND60: "human-sized buys" filter for unhooked v4 fee-100 pools (analyst1, 9-10 Oct)

Lens: behavioural fingerprint of the buyers (who is on the other side). Scripted bots buy random-looking amounts; people type round numbers (0.01, 0.05, 0.1, 0.5 ETH). If the first minute already contains human-sized buys, a real crowd is there and the operator is less likely to pull at once.

Data: P2.pkl (r6weth tape, 21 Sep - 7 Oct), u100 = v4 fee 100, unhooked hook suffix not 144. 1,336 enterable pools (reserve >= 0.5 ETH at +60 s), sim from sim2.py (lag 3 s, impact by Kyle lambda, fee 1%/side, pull-aware: pulled before exit = 0).

Feature (known at +60 s): share of buys (>= 0.0005 ETH WETH in) in the first 60 s whose amount is a multiple of 0.01 ETH (>= 0.01). Needs >= 3 buys.

| bucket (u100, entry +60 s, tp3) | n | mean x per $1 | zero |
|---|---|---|---|
| no round buy | 190 | 0.60 | 77% |
| < 1/3 round | 819 | 0.92 | 68% |
| 1/3 - 2/3 | 167 | 1.25 | 57% |
| > 2/3 | 160 | 1.40 | 51% |

Rule: round share >= 0.5 -> n=263, tp3 1.37 / tp5 1.53 / tp10 1.51. Halves (by time) tp3 1.26/1.47, tp5 1.35/1.71. Ex top 5 trades tp5 1.46. Holds inside swap-count bins (8-20 swaps: 0.98 vs 0.72; 20-60: 1.09 vs 0.63 for >=2 vs <2 round buys). Correlation with swap count 0.07, with ping-pong flip rate -0.05: independent of BUSY60/PERBLK style filters (not yet tested together).
Not in other families: v3 fee100 (1.06 flat), hooked (2.2-2.4 flat), dyn-fee v4 (0.5-0.7, loses).
Also: price-sign flip rate (Roll style wash proxy) top quartile u100 0.59 (n=321), others 1.0-1.2: avoid filter, adds nothing once round filter is on (round & flip<q75 1.58).

Caveats: threshold 0.5 chosen after seeing the monotone table (in-sample); tape ends 7 Oct; mean 1.55 is modest next to hooked 2.3. Not a replacement, a diversification arm (different mechanism).
Forward test rule (for testmgr): v4 fee-100 pool, hook suffix not 144, at creation+60 s: among buys in first 60 s (WETH/ETH in >= 0.0005), share that are multiples of 0.01 ETH within 0.0002 ETH >= 0.5 with >= 3 buys, and reserve >= 0.5 ETH -> buy, tp5 (sell on 5x), else sell at 20 min or at pull. Kill: forward mean < 1.0 over n >= 60.
Script: y1-y3.py, y11.py in /opt/crew/work/analyst1.