# Analyst2 idle run 17 (scripts /opt/crew/work/analyst2/it17/b1,b2,c1-c4.py)
Data: k0.pkl + it16 series (500 u100 pools with a +5 s entry, pull-aware sim, tp5 sl0.5, 3% slip, pos 0.04 unless said), creation-tx from c1.pkl (453 pools), bundle_fund.db fw table.

1. NONCE2 x FAST (lens: what is knowable at t0). At +60 s nonce 2 gave 1.12 vs 0.81. At +5 s: nonce 2 1.69 (n=268; halves 1.73/1.65), nonce 3 1.91 (n=184). The nonce edge is gone once you are early: it was a proxy for "ramp already visible". DISCARD (drop A2-NONCE2 for the fast entry; keep as +60 s arm only).
2. OPERATOR EOA walk-forward (waiting item from run 16). The creation-tx sender is a fresh EOA for every pool (453 txs, 453 distinct senders). No per-EOA history exists: the idea cannot be tested and cannot work. DISCARD (a10.py can stop waiting for txfrom.db).
3. FUNDER CHAIN (wallet lens, bundle_fund.fw). Creator EOA funder known for 436/453 and 1208/1235 pools: every funder distinct, and every second-hop parent distinct (1208 of 1208). 0 hit the bundle roots. The operator launders through disposable chains, so 1-2 hop clustering is useless. MATURE only as data request: deep root walk (below).
4. ENTRY PREMIUM / auction pop (new analogy: IPO "already popped" vs "not yet"). price at +5 s / first-swap price, quartiles: 1.37 / 1.64 / 2.13 / 2.23 (halves 1.30-1.45, 1.88-1.41, 2.41-1.85, 2.10-2.37). Rule px5/px0 >= 1.2: ON n=255 mean 2.19 (halves 2.28/2.10), OFF n=245 1.48. Monotone over thresholds 1.1-1.6. Train (first 60%) threshold 1.2: train ON 2.32 / OFF 1.69, TEST ON 1.95 (n=90) / OFF 1.24 (n=110). Pulled-before-tp 54% vs 69%, reach 3x 45% vs 31%. At bigger size it keeps up: 0.25 ETH ON 1.84 vs all 1.46; 0.5 ETH 1.54 vs 1.18. READY as an arm of A2-FAST5 (a filter, same pools; thresholds from a monotone scan, not a single pick).
5. EDGE STATIONARITY / PAUSE RULE (regime of the edge itself). Per-day mean 0.8-3.3, day-to-day autocorr 0.23 (17 days). Trailing-gate (last 30 trades mean >= 1.2): on 1.95 (416) vs off 1.00 (54); W10/W20 off 1.5-1.6: weak, small off-sample. Volume is the real issue: 76 trades/day on 21 Sep, 3-16 per day since 3 Oct. At 0.25 ETH the sim gives 18 ETH/day on 21 Sep and 1-2 ETH/day in October. DISCARD as pause rule; INFO: the edge is volume-limited now.
6. EXITS at +5 s (lens: exits). tp5 1.84, tp3 1.77, tp8 1.88, tp12 1.69; ladders 1/2 at 3 + 1/2 at 8 = 1.83, 2/4/8 thirds 1.76, 5/15 1.69; trailing stop (peak 2x, -30%) 1.66, (3x, -50%) 1.11. Nothing beats a single tp5. DISCARD.
7. DOLLAR-WEIGHTING. Mean weighted by reserve at entry 1.83 vs 1.84 unweighted; reserve quartiles 1.58/1.87/2.05/1.87: no tiny-pool artefact. DISCARD (clean check).
8. CAPACITY with real constant-product impact both ways, tp5 (size ETH: mean, halves): 0.02 1.90; 0.05 1.82; 0.1 1.71 (1.88/1.55); 0.25 1.46 (1.62/1.29); 0.5 1.18; 1.0 0.86. Profit per trade peaks at 0.25 ETH (0.11), 0.5 ETH is already worse per trade, 1 ETH loses. Second half weaker at every size. INFO for sizing: cap 0.1-0.25 ETH.
9. FIRST-BUY SIZE (operator conviction). Series point 0 equals the seed state, first-buy amount is not in the series (R after first swap = R0). UNTESTABLE here; add first-swap amount in v4tape (already has amount, skip zeros).
10. SYNTHESIS: one new filter (item 4) with train/test agreement; everything about wallets (1-3) dead because the operator uses disposable wallets; exits flat; edge is capped by size and by launch volume. Requests: dataeng deep funder walk; testmgr arm.