# Analyst2 idle run 25 (10 Oct 2026): 10 iterations, 2 arms for testmgr, 1 data request
Data: r6nat.db (native tape to 9 Oct 04:01 UTC, 2,114 pools), live v4tape forward (9 Oct 07:49-23:20 UTC), r6weth u100 pickle. Scripts: research-box /opt/crew/work/analyst2/it25/ (base.py loader, a1..p2). Sim as before: entry first swap +5 s, 0.1 ETH, stop 0.5, 3% slip, fee both ways, bag = 0 if the pool is pulled first, TP needs reserve >= 2 x position x multiple.

## Results
1. DAY-CLUSTERING (new lens: hit rate is not iid). Native A1 daily hit rate 0.04-0.62, variance 4.2x binomial (permutation p=0). Yesterday does not predict today (corr 0.05), but a trailing-10-resolved-trades gate (mean >= 1.0) splits A1 1.42 (n=616) vs 0.78 (n=641), halves 1.44/1.40 vs 0.74/0.82. FORWARD (132 pools 9 Oct): on 1.32 / off 1.43, off n=23: does not reproduce. MATURE, no arm. Forward A1 is better than history (1.41 vs 1.09), so the gate rarely closes.
2. LIMIT-DIP ENTRY (new lens: entry type). Limit buy at 0.95/0.9/0.8 x the +5 s price inside 30/120 s: fills 0-5% of pools (ramped pools only go up). DISCARD.
3. NAME ECHO of recent winners (word overlap with a winner of the last 6/24 h, 1,582 named pools): win rate 0.406-0.424 vs 0.401; A1 1.16 vs 1.07. DISCARD.
4. PREDICTABILITY CEILING (walk-forward logit and small GBM, 10 features known at +5 s, train first 50/65%): A1 OOS AUC 0.63-0.69 (mostly the dead full-range kit; without it AUC 0.61-0.63, top half 1.56-1.62 vs 1.32-1.42). u100 WETH AUC 0.52-0.57 (premium alone 0.60). H3000u AUC 0.94: driven by swap count before entry and initial liquidity (see 5). INFO: ceiling is low for A1/u100.
5. H3000u RAMP-PRESENT FILTER (from 4). Native unique-hook fee-3000, tp5, entry +5 s: swaps already seen before entry n_pre>=3: mean 4.09 (n=180 of 243), halves 3.79/4.39, zero-rate 14% vs all 3.31 / 26%; n_pre<=2: 1.08 (n=63, median 0, 57% zero). Monotone in the threshold (>=2 3.77, >=3 4.09, >=4 4.27, >=5 4.39), so not a picked cutoff. 95% bootstrap CI of the filtered mean 3.81-4.35. Forward (9 Oct, n=11): n_pre>=3 3.47 (n=9) vs <=2 0.95 (n=2). READY as an arm of the H3000u native test.
6. A1 FULL-RANGE MINT (first mint tick_upper at max): 383 of 1,268, 0.53 mean, 80% zero. 148 of them are the known dead price class (0.04); the other 235 0.84 vs 1.36 for non-full. Duplicate of the dead-kit finding (run 22); forward only 4 pools. DISCARD (zero-latency rule, but no new money).
7. u100 predictability: see 4. DISCARD.
8. LP-ADDER IDENTITY (who sends the first liquidity add; earlier runs looked at swap senders and at tick_upper only). Table over native families: 0x58daec31 (the operator: 1,490 pools, A1 1.02, H3000u 3.46, hooked fee100 0.05) and ONE other sender 0xd0fd42bd8374ac659d3d81242d032211c2481055: 66 fee-100 unhooked pools, mean 2.46 at tp3, 4.16 at tp5. r6weth u100: one sender only.
9. D0FD DEEP DIVE (see arm 2 below). Same-day control (other A1 pools on the 12 days it was active) 1.01 vs 2.46, so not a day effect. Template: first mint L=1.67e22 (also 5e22, 4.45e21), tick_upper ~198k, price 2.8e-9; pulled in 98% after a median 4,232 s; peaks 6-85x. Latency tolerant: +60 s entry tp5 3.75. Depth: only 0.1 ETH fits (pos 0.25: 1.66, 0.5+: 1.49 because tp needs reserve >= 2 x pos x multiple). All 54 tp5 hits had a prior print >= 3.5x (gradual, no jump). Bursty: 26 Sep 11, 29 Sep 14, 30 Sep 15, 1 Oct 9, then 1-4/day, 8 Oct 4, 9 Oct 2. Forward 9 Oct: 1 pool (0xe92bb932.., 17:26 UTC) hit 4.7x at tp5 in 20 min.
10. DATA REQUEST + synthesis: see below. Bridge ETH inflow and stable mints already killed by analyst1 (iteration 1 of its run 14-ish), not repeated.

## Arm 1: A2-H3000U-RAMP (add to the H3000u native test)
Pool: v4, quote native, fee 3000, hook used by <= 2 pools (unique hook). Entry: first swap + 5 s, 0.1 ETH, only if >= 3 swaps were already seen in the pool before the entry. Exit: sell at 5x entry price (reserve >= 2 x pos x 5), stop 0.5, 4 h time stop, pull = loss. Baseline = same pools without the filter (3.31). Kill: filtered mean not above unfiltered over 40 trades, or n_pre<=2 pools beating >=3 pools.
## Arm 2: A2-D0FD (new test)
Pool: v4, quote native, fee 100, no hook, first positive liquidity add sent by 0xd0fd42bd8374ac659d3d81242d032211c2481055. Entry: first swap + 5 s (up to +60 s ok), 0.1 ETH max (depth). Exit: sell at 5x (second arm: 3x), stop 0.5, 4 h time stop, pull = loss. Evidence: n=66, tp5 mean 4.16 halves 4.00/4.32, tp3 2.46 halves 2.88/2.04, same-day control 1.01, forward 1 of 1 hit. Kill: under 1.5 over the next 25 trades, or outsiders cannot sell at 3-5x (outside-money test pending, see request).
## Data request (ppdata/dataeng)
See fw/a2-req-d0fd.md.