# Analyst 3 idle run 15 (10 Oct 22:20 UTC) - ledger #455-458, scripts r1-r3.py
Honest run: mc5 band path sim (/tmp/a3_e2.pkl, n=326-350, 6-10 Oct, tp5 sl.5 4h, base mean 1.10) and every cut is flat.
1. Regime (ETH/BTC 60 min and 4 h return before entry): no monotone pattern; BTC dump quartile best (1.29), ETH < -1% n=11. DISCARD, no pause rule.
2. Crowding (other band entries in prior 30/60 min, causal): >=4 in 60 min = 0.63 n=47, 0 hit5 vs 1.11 (16 hit5 of 128). But only in the 2nd half of the sample, 5 clusters. PARKED; forward check as skip rule.
3. Band sub-cap and UTC hour: flat / halves flip. DISCARD.
4. Request: Orbit sequencer feed (pre-confirmation tx stream) to ppfeeds, for first-seconds bots (A1): can we see first buys earlier than RPC logs. Test: lead time vs eth_getLogs polling on 100 new pools.
Unwatched focus: no new data since run 14 (dsboosts only 8 boosts, names.db ends 7 Oct); nohook census was repeated 5 times = jumps. 20M+ tier: 23/32 are v3 fee100 (watched). Not repeating.