# fmmc5 and market mood: analyst2 cross-check (9 Oct 2026)
Data: A1 replay /root/fwa1/rug/now_trades.csv (201 closed trades, 13 Sep - 8 Oct; x = exit multiple NET of fee+gas both ways (bt.py); corrected 9 Oct per tm-audit-194), BTC/ETH 1-min closes /opt/ppcopy/prices.db (ends 7 Oct 19:40 UTC, so 192 trades have a price). Scripts m1.py, m2.py in /opt/crew/work/analyst2.

1. Base rates: 20% of trades end >= 1.0x, 18% >= 2x, median result 0.48x (the stop). Mean 2.0-2.9x is carried by a handful of runners (6 Oct: 7 trades, mean 12.7x).
2. 8 Oct (15 losses in a row) is NOT a freak. With a 20% win chance and about 200 trades, the longest losing streak is 18 in the median simulation (p90 26, P(a streak >= 15) = 79%). The replay itself has streaks of 14, 14, 12, 11 and four of 8. So 10+ loss streaks happen several times a month; they say nothing about the rule being broken.
3. Market mood, entry-time moves (n = 192): BTC 24 h change down >1%: 35 trades, 11% >= 2x, mean 1.34x; flat: 123 trades, 20%, mean 2.38x; up: 34 trades, 18%, mean 2.53x. ETH 24 h: down 14% / flat 20% / up 14% (not monotone). 6 h moves: BTC up 15 trades, 7%; down 14%. Rank correlation of result with BTC/ETH 6 h and 24 h changes: +0.09, +0.08, 0.00, -0.09. The gap on BTC-down days (11% vs 20%) is about 1.3 standard errors with n = 35. NOT significant, not monotone for ETH.
4. Day level: 22 trading days, hit rate by day 0-60% with 1-26 trades per day; the worst days (28 Sep 0 of 11, 8 Oct 0 of 2) have BTC -1.7% and -2.7%, the best (22 Sep, 6 Oct) BTC +3.8% and -0.3%. No pattern.
5. Verdict: no usable on/off radar from BTC/ETH. Any rule built on this would be fit to 3 weeks and about 35 down-day trades. If Ben wants one anyway: pause on days when BTC 24 h < -2% (only 8 Oct and 24 Sep qualify, 13 trades, mean 1.4x); it would have saved little and cost 2.56x-style runner days. Better than a mood switch: size by loss streak is useless too (outcomes are independent draws at 20%).
6. Metas: the A1 replay has no meta label per day; kit A (24-30 Sep) vs kit B (since 1 Oct) is the visible regime and the hit rate is similar (A ~18%, B ~16% >= 2x). Not tested further (needs the launch-count series from ppdata).
7. Size: at 20% hit and median 0.48x, the P&L is a lottery on a few 10x+ runners (top 2 trades carry about 1/3 of profit per A1's own table), so a bad stretch of 15-20 losses is expected at any size. $10 is a sensible size until there are 100+ live trades; going to $50 needs live mean >= 1.5x after fees over 100 trades with the top-2 removed still > 1.