# a1 iteration 45 (11 Oct): LP pull survival and first-300s order flow on v4 nohook (swapamt.db + liq_owner.db + canon, scripts o1/f1-f8.py)
Universe: born-small (cap_first<150k) v4 hooks=0 fee100 pools 21 Sep-9 Oct (weth 2637, native 790); outcome = canon max_cap>=1M (canon stale 9 Oct).
1. First-300s order flow (new swapamt.db): buys count/volume/median buy size monotone with P1M (weth nb>=72 4.0% vs 0.1-0.4%; native 6.6% vs 0-0.7%), seller share and top-buy share likewise. Pure popularity; within nb>=40 nothing separates (top share, sell share flip across halves). Known family (#421), needs pull-aware sim (#420: 1.03). DISCARD.
2. LP pull is the whole story: fee100 pools with ANY LP remove in first 15 min: weth 2/2271 (0.1%) reach 1M, native 0/617. No remove in 1h: weth 33/366 (9.0%), native 19/173 (11.0%). Removes <60s 583 pools, 60-300s 843 (weth). Pulled pool = dead; runner = pool whose LP is never pulled.
3. But pull is not forecastable: survivors at +300s with nb>=40 still 67% pulled later (weth 756/1118, native 129/226), P1M 3.1% weth / 8% native. Owner novelty (96% of owners 1 pool), add size, single add (no variation), NFT burn = the pull itself. No predictor. DISCARD as entry filter. Pull = rug by design in the kit; the 14% survivors are rare organic coins.
4. Wash index (buys matching a sell size within 300s) and repeated buy sizes: flat/confounded by popularity. DISCARD.
5. USDG/SUSD nohook (7491 born-small pools, 96 hit, hundreds of custom fee tiers 50 to 991200 ppm): hit rate 1.3% overall, 3% at nb>=10, no fee tier separates. DISCARD.
Implication: a nohook weth/native strategy can only work if it exits before pull (impossible: pull is a single tx, no mempool on the sequencer) or selects the 14% never-pulled coins. Nothing found to select them at +300s. Open: tx.from of pulls (#1247 part 2); extension of liq_owner to all fees/quotes (request posted to ppdata).