🔬 Multi-Timeframe Battle: Jan 2025 – Jun 2026

128 backtests | 16 windows × 4 variants × 2 capital models | Identical v7 logic

Brutal finding: ALL variants lose in the 2025–now period.
No timeframe or pair count flipped the sign. The SAR cluster edge that worked in 2023 has degraded under 2025's event-driven, de-correlated market structure.

Summary — Unlimited Capital ($100 fixed stake)

Variant	Trades	Tot PnL	Avg/Window	Win+ / Loss− / 0
1m_12_unlimited	181	-1,268.0	-79.2	+5/−10/0=1
15m_12_unlimited	169	-1,277.9	-79.9	+4/−12/0=0
5m_20_unlimited	281	-2,634.6	-164.7	+3/−11/0=2
15m_33_unlimited	373	-3,061.1	-191.3	+4/−8/0=4

Summary — Realistic Capital ($500 fixed stake)

Variant	Trades	Tot PnL	Avg/Window	Win+ / Loss− / 0
1m_12_realistic	151	-5,873.9	-367.1	+4/−11/0=1
15m_12_realistic	154	-6,533.2	-408.3	+4/−12/0=0
15m_33_realistic	121	-7,243.7	-452.7	+4/−8/0=4
5m_20_realistic	142	-8,841.0	-552.6	+3/−11/0=2

What This Means

Pair count > Timeframe. 15m/12 lost ~60% per window of what 15m/33 lost. Shrinking the universe to core movers mattered more than going faster.
1m/12 was the least-damaging. It had the highest window win rate (+5/−10) and smallest avg loss (−79 unlimited). But it still lost.
5m was the worst. It caught the most noise without enough resolution benefit. The "middle speed" trap.
The edge is gone in 2025. Not because of timeframe, but because crypto stopped moving in macro-correlated clusters. ETF flows, single-coin events, and AI rotations broke the 2023 coherence that PCA could capture.