Meta-Controller Experiment

All Strategies Tested (2025–Now)

Strategy	Trades	Tot PnL	Avg/Win	W/L
micro_volume (Least bad individual)	57	-24.0	-1.5	8/8
micro_rsi (Low trade count)	16	-33.0	-2.1	8/8
micro_session (Most trades, most losses)	69	-56.3	-3.5	8/8
micro_ema (Churns trades, worst net)	80	-62.8	-3.9	8/8
Momentum Controller (Picks recent winner)	42	-39.2	-2.4	8/8
Contrarian Controller (Picks recent loser)	61	-45.1	-2.8	8/8

What The Controllers Chose

Momentum controller: pick strategy with best cumulative PnL in prior 2 windows.
Contrarian controller: pick strategy with worst cumulative PnL in prior 2 windows.

Window	Momentum Pick	PnL	Contrarian Pick	PnL
20250101-20250302	micro_session_unlimited	-5.5	micro_rsi_unlimited	-10.8
20250131-20250401	micro_session_unlimited	-21.2	micro_rsi_unlimited	-19.7
20250302-20250501	micro_session_unlimited	5.8	micro_rsi_unlimited	10.3
20250401-20250531	micro_rsi_unlimited	25.0	micro_ema_unlimited	17.5
20250501-20250630	micro_rsi_unlimited	12.7	micro_ema_unlimited	10.0
20250531-20250730	micro_rsi_unlimited	13.0	micro_session_unlimited	10.2
20250630-20250829	micro_rsi_unlimited	4.3	micro_ema_unlimited	0.8
20250730-20250928	micro_volume_unlimited	-3.9	micro_ema_unlimited	-2.7
20250829-20251028	micro_volume_unlimited	1.4	micro_session_unlimited	3.0
20250928-20251127	micro_ema_unlimited	-17.5	micro_session_unlimited	-19.6
20251028-20251227	micro_volume_unlimited	-24.0	micro_rsi_unlimited	-22.5
20251127-20260126	micro_volume_unlimited	-6.0	micro_session_unlimited	-3.6
20251227-20260225	micro_rsi_unlimited	-27.0	micro_ema_unlimited	-25.4
20260126-20260327	micro_session_unlimited	-22.0	micro_ema_unlimited	-21.7
20260225-20260426	micro_session_unlimited	10.3	micro_rsi_unlimited	15.0
20260327-20260526	micro_volume_unlimited	15.6	micro_session_unlimited	14.1

Lessons for a Live Meta-Controller

The controller cannot manufacture edge. It only redistributes capital. If every component is −EV, the portfolio is −EV.
Mean-reversion of strategy performance: A strategy that wins 2 windows in a row is more likely to lose the third. "Picking winners" just chases luck.
Trade density matters: Each micro-strategy only generated 1–5 trades per window. A controller needs > 30 trades to estimate Sharpe meaningfully.
The live vision is still valid, but only after at least ONE component is +EV. The controller's job is capital allocation, not alchemy.
Most promising raw edge found so far: micro_volume at −$0.42 per trade. An adaptive stop or session filter might be enough to flip this.

🦾 Meta-Controller Experiment

All Strategies Tested (2025–Now)

What The Controllers Chose

Lessons for a Live Meta-Controller