# analyst1 iterate run 20 (10 Oct 2026): 10 iterations, nothing ready for testmgr
Scripts da1-da10.py (+ da3f/da6f fetchers) in /opt/crew/work/analyst1. Sim = P2 tape (r6weth, to 7 Oct), entry +60 s, tp3, fee/impact model of sim2.py.
New lenses: pair token ordering (address vanity/orientation), attention cannibalization, chain congestion (base fee, block fill), round-cap stall levels, whale-buy share, bytecode metadata fingerprint, organic-venue baseline, hook skim audit (real transfers vs swap events), closing-line-value drift as edge-health gauge, concurrent-launch density.

1. TOKEN ORDER (token is token0 vs token1): only 6% of pools have token as token0 (WETH sorts lower for nearly all). u100 0.99 vs 0.98, h100 2.10 vs 2.28 (n=21), v3100 0.49 (n=33) vs 1.12. Thin, no mechanism. DISCARD.
2. CANNIBALIZATION (rank by reserve among pools created in prior 5 min, at +60 s): u100 top rank 1.09 vs not top 0.94 (n=62/189), h100 2.24 flat, v3100 1.30 vs 1.09 (n=44/87). Concurrency count 0 vs 3-6: flat in all three families. DISCARD.
3. CONGESTION (creation block base fee, gas used share, tx count): chain sits at the 0.02 gwei floor; terciles u100 0.99/0.86/0.95, h100 2.54/2.64/2.46, v3100 0.99/1.03/1.07. No regime from gas. DISCARD.
4. ROUND-CAP STALLS (peak cap of e1 runner/control sample n=304): below/above $1M (0.9 band) 9/15, $2M 4/9, $5M 1/3, controls non-round same shape. No resistance at round caps, no exit rule. DISCARD (sample thin).
5. WHALE BUY (largest buy in first 60 s as share of reserve / of buy flow / ETH): u100 quartiles 1.08/0.91/0.91/1.04, h100 2.40->2.15, v3100 1.08/1.15/1.19/0.99. Flat. DISCARD.
6. BYTECODE METADATA (solc version, IPFS hash of CBOR tail, size, eth_getCode of 843 pool tokens): ALL tokens of a family are byte-identical clones: u100 2 hashes (399/15), h100 1 hash (122), v3100 2 hashes (188/36+). Code carries no per-coin information; all the information is in who runs the pool. INFO + DISCARD.
7. ORGANIC VENUES (v3 1% and 0.3% pools, n=468 pools, 48-69 enterable): best cell mean 1.27 at +10 s tp3 1h, but 1.0 +-0.1 elsewhere and median 0.5-1.0 for tp>=3; no cell survives fees. Organic pools = efficient coin flip; edge lives only in operator families. DISCARD (useful baseline).
8. HOOK SKIM AUDIT (new tool): v4 Swap events vs ERC20 Transfer logs in the same tx, 238 early swaps (209 u100, 29 h100) token side ratio 1.00 exactly; 221 h100 sells quote side: transfers = amount (2 hops, 2.00 = full delivery, no skim). The 0x144 hook (afterSwapReturnsDelta bit set) takes nothing on those swaps, so model out/model ~0.99 holds. INFO (positive for sellability).
9. CLOSING LINE VALUE (sports-betting translation: mean log move in 30 s after entry as an edge-health number, per day): same-day corr with payoff -0.37 (h100, 12 days) / +0.23 (u100) / +0.53 (v3100); next-day -0.75/+0.15/-0.14. Inconsistent signs, n=12-17 days. DISCARD as gate; health must stay on payoff.
10. CONCURRENT LAUNCH DENSITY on e1 sample: more launches in prior 10 min = better (4.2x vs 2.5x) opposite of cannibalization, but e1 is runner+control sample, rates are not population. DISCARD, not testable on it. DATA REQUEST: ppfeeds round-trip quote feed (below).

## Data request (ppfeeds)
fw/a1-iter20-ppfeeds.md. Forward feed `roundtrip.jsonl`: for every pool the family feeds already emit (u100, h100, v3100, v2 seed-1, native H3000), at entry age +5 s and +60 s run an eth_call dry-run (stateOverride, PoolManager unlock/router, no money): buy 0.1 ETH, then sell the tokens received. Fields: pool, family, age_s, eth_in, tokens_out, eth_back, rt_ratio, error. Test: median rt_ratio vs sim model (0.98 u100/h100, 0.93 v3 with 1% fee); kill = rt_ratio < 0.93 or error > 5% in a family. It replaces the "only a real $5 buy+sell proves sellability" gap for paper picks.