Rotation idea bench, 3 Oct 2026 evening (Bangkok). 78 ideas on 14 point in time CoinMarketCap baskets + majors, Jan 2023 to Sep 2026.
Each idea: trade stats (A = 2023-24, B = 2025-26) and the 3 slot book (BUY day entries only, swap extended half into fresh BUY,
0.5% cost per side). Files: x50.py (bench), x50_run.py / x50_run2.py / x50_run3.py (ideas), x50_grid.py (168 neighbour settings),
x50_round*.txt / x50_grid.txt (outputs), rule4_v42.py (live rule after the change).
Kept (live as rule v4.2, tested baskets only):
 1. HOT MARKET window: when 50%+ of other baskets are risk on, a close above the 20 day high is enough (30 otherwise).
 2. Breadth gate 35% (was 25%).
 3. Trailing stop: sell what is left on a close 20% under the best close since buying.
 Book: +1006% (worst drop -25%) vs v4.1 +511% (-33%); per year 2023 +127/+107, 2024 +175/+119, 2025 +16/-1, 2026 +51/+37.
 161 of 168 neighbour settings (trail 0/15/20/25, breadth 25/30/35, hot market 40/50/60%, fast window 20/25, rank r30/r7) beat v4.1 in both halves.
Killed (examples): holding the strongest baskets every day (state books, 2025 -30 to -70%), dual momentum, EMA trend, ensemble Donchian,
 pullback entries, bigger breakouts (60/90 day), RS vs BTC entry filter, volume filters, squeeze, two closes, time exits (5/8/12 days),
 inverse volatility sizing, idle money in BTC or majors (bigger drops), 2 slots (bigger drops), ranking changes (noise).
