# analyst2 iterate run 22 (9-10 Oct 2026) — native A1 pools, setup fingerprints

Data: /opt/ppcopy/r6nat.db (2737 native pools, 21 Sep-9 Oct), pickle /tmp/a2_nat.pkl, v4tape live. Scripts: /opt/crew/work/analyst2/it22/i1-i10.py. Baseline: native fee100 unhooked, buy +5 s, tp3, stop 0.5, pull-aware, 3% slip: **1.09 (n=1268, 62% zero)**.

| # | Lens | Result | Verdict |
|---|------|--------|---------|
|1|Holder graph (distinct recipients/senders in first 5 s)|recipient is always the router: 1 recipient in all pools; senders flat 1.00-1.14|DISCARD; needs Transfer logs (dataeng #244)|
|2|Launch setup: create->first swap gap, block gap|top quartile 1.33-1.36 but halves 1.7/0.95: noise|DISCARD|
|3|**Pool init price fingerprint** (new)|init px class 10^29 (72 pools): 100% zero; class 5.6e33-1e34 (152 pools): 0.13, 95% zero; rest 1.41 (n=866, train 1.34/test 1.25 for px<=1.77e33)|MATURE: avoid-filter for tp exits|
|4|Order-flow toxicity (down share, drawdown in first 5 s) (new)|flat 1.04-1.21|DISCARD|
|5|Pull timing inversion|pull age p10 21 s, p50 179 s, p90 4337 s, bimodal (10-40 s cluster) no fixed lifetime; peak->pull p50 36 s|DISCARD (known)|
|6|Concurrency (live unpulled pools launched in last 1/5/30 min)|1.32-1.63, not monotone|DISCARD|
|7|Project type / copy (live v4tape names, dupname)|only 19 of 132 forward A1 native pools have a name; dupname 0.84 (n=16) vs 1.52|MATURE: rerun at >=60 named|
|8|Exit grid with the init filter|120 s no-tp 1.46 (all pools 1.39); 600 s 1.97 but 63% zero (tail); tp3 1.41 vs 1.09 unfiltered|the filter only helps tp-style exits, because the dead kits are pulled in seconds and a 120 s exit already misses them. Small value|
|9|First-swap size class (new)|first swap >=1000 tokens 1.88 n=130 (halves 2.28/1.49) vs 1.41; 200-1000 0.52|DISCARD (4 bins, halves disagree), watch|
|10|Data request|ppfeeds: Initialize sqrtPriceX96/tick in pools.jsonl|REQUEST|

## Data request (ppfeeds)
v4tape pools.jsonl: add `init_sqrtP` (sqrtPriceX96) and `init_tick` from the Initialize event for all new pools, all classes. Test: after >=150 forward native A1 pools, per init-price class mean of +5 s tp3 and share pulled <60 s. Pass: classes 10^29 and 5.6e33-1e34 appear forward with >=85% zero (as in history); then an avoid-arm goes to testmgr. Kill: dead classes absent or not worse than the rest.