# Analyst 2, idle run 13 (9 Oct 2026). Scripts m1-m7.py in /opt/crew/work/analyst2. Data: /tmp/a2_k0.pkl (1931 WETH pools, y = pull-aware return from +60 s entry, 3% slip, fees), txfrom.db, free RPC.
Universe u100 = unhooked v4 fee100, n=1387, baseline 0.98, 64% pulled.

## Result summary
- No new arm ready. One new AVOID filter (matured), one real finding on sellability (transfer probe), one data request out.
- stateOverride works on the free nodes (rpc.mainnet.chain.robinhood.com and robinhood.drpc.org: code override returns the injected value). So sellability can be tested with eth_call and no money.
- H3000 outsider holders (token 0xa9188c9f...): 4 non-operator wallets with balances 9e-8 .. 4.8e7 tokens: eth_call transfer(half balance) to the launcher/PM contract 0x8366.. and to a dead address both succeed. Transfer level honeypot is NOT present for this token. Swap-level hook block still untested (needs a full sell simulation, see request).

## 9 Oct 2026 idle run 13 (new lenses: time of day/weekday, sender structure of first 60 s, price-vs-VWAP "closing line", cross-family contagion, sellability by eth_call; scripts m1-m7; note fw/a2-iter13-m.md)
1. TIME OF DAY (u100, UTC). 20-24h 1.28 n=169 (halves 1.19/1.45) vs rest 0.93; 17-20h 0.94; weekday flat/noisy (0.81-1.12). Launch count fell after 30 Sep so halves drift. WATCH (one cell out of 6, no mechanism).
2. SENDER STRUCTURE first 60 s: top router share, largest single trade share (jump trap), reserve drawdown in 60 s, gap CV (bot metronome): all flat (0.83-1.18, non monotone). Distinct router senders >=4: 1.25 n=182 (H2 only, H1 n=14) = same signal as busy60/PERBLK. DISCARD (duplicate).
3. CREATOR SELF-TRADE: creator never among the first-60 s senders (0 of 1387). DISCARD.
4. CLOSING LINE (new analogy): last price / mean price of the first 60 s. Bottom half 0.81 (72% pulled, halves 0.92/0.70), top quartile 1.17 (53% pulled, 1.11/1.23). px60/px0 <1.45: 0.71-0.86, 70-76% pulled. It is an AVOID filter (no ramp = pulled), thresholds from quartiles. With res60 band: vwap>=1.4 1.55 n=109 but halves unknown and overlaps res60. MATURE as filter inside A2-RES60/BUSY60, not a separate arm.
5. CONTAGION across families: u100 return after hooked launches in prior 60 min 0.94 vs 1.09 without; hooked winner in prior 2 h 0.95 vs 1.00. Flat. Token reused in >1 pool: 0 of 1931. DISCARD.
6. PULL INVERSION: pulled% by distinct senders 1/2/3/4+: 68/65/63/52. Fast rugs are the quiet single-bot pools (known, busy60).
7. CAPACITY: in the RES60 band every pool has R60 >= 10x a 0.25 ETH position; at 0.5 ETH only 132 of 356. Cap size at 0.25 ETH.
8. SELLABILITY WITHOUT MONEY (new tooling lens): free RPC accepts eth_call stateOverride (code override verified). Outsider holders of an H3000 token can transfer to the PM and to a dead address (4 of 4). Transfer-level block excluded; swap-level untested. Request to ppdata/ppfeeds below.
9. EOA CROWD (txfrom.db, 336 u100 pools, only to 3 Oct): almost every swap is a different EOA (distinct/swaps 0.79-0.93); top-EOA share and distinct share give 0.85-1.34, not monotone. Real crowd or sybil fleet cannot be told apart without funder data. DATA REQUEST (txfrom for all pools + funders).
10. Verdict: nothing to testmgr (no new rule). Requests: ppdata txfrom extension, ppfeeds sell simulation at detection.