🔁 BTC Vol Regime + Old Trend Filters

64 backtests | 16 windows × 4 variants | $100 fixed stake | Jan 2025 – May 2026

Old filters make the new strategy WORSE.
The 1h gate delays entries on winners more than it blocks losers. SAR trend filter is redundant when every window's 1h bias is already bullish.

Aggregate Results

Variant	Trades	Tot PnL	Avg/Win	Win/Loss
Base (no filter)	16	-27.1	-1.7	+8/−8/0=0
+ Gate	16	-37.3	-2.3	+8/−8/0=0
+ SAR Trend	16	-38.8	-2.4	+8/−8/0=0
+ Gate + SAR	16	-43.9	-2.7	+8/−8/0=0

Per-Window Gate Impact

Window	Base PnL	Gate PnL	Delta
20260327-20260526	15.6	7.3	-8.3
20250501-20250630	11.1	4.6	-6.5
20260225-20260426	21.4	15.7	-5.7
20250302-20250501	6.5	3.2	-3.3
20250829-20251028	5.6	2.4	-3.2
20250928-20251127	-21.3	-23.9	-2.6
20250101-20250302	-11.2	-12.7	-1.5
20250730-20250928	-3.9	-4.7	-0.8
20250531-20250730	12.2	11.4	-0.8
20251127-20260126	-4.7	-5.1	-0.4
20250630-20250829	3.6	3.8	+0.2
20251227-20260225	-28.0	-27.8	+0.2
20251028-20251227	-20.9	-20.6	+0.3
20250401-20250531	24.1	24.7	+0.6
20250131-20250401	-17.6	-15.7	+1.9
20260126-20260327	-19.6	-0.0	+19.6

Key Findings

Gate saved money once: Jan–Mar 2026 (saved +$19.6 by blocking a −$19.6 loser).
Gate cost money three times: Mar–May 2025 (−$6.5), Feb–Apr 2026 (−$5.7), Apr–May 2026 (−$8.3). In each case it delayed or missed a strong winner.
1h gate is too slow for 15m entries. By the time the 1h BTC gate flips from closed to open, the 15m squeeze breakout has already happened.
SAR trend is always bullish. On every single window, BTC 1h SAR was above price at some point within the window, so the filter never actually blocked a trade. It was decorative.
Conclusion: The old v4 macro stack (gate + PCA/SAR) was built for multi-pair, multi-day regime detection. It does not translate to single-pair, intra-day vol-regime trading.