Regime-aware noise filters for crypto perpetual signals

Crypto perpetual markets move through regimes faster than most fixed-parameter systems expect. A threshold that is selective during quiet liquidity can become reckless during a volatility expansion, while a filter that avoids chop can miss trend continuation after forced deleveraging.

We model regimes as observable state rather than as labels discovered after the fact. Volatility percentile, spread behavior, funding pressure, and volume participation all contribute to a lightweight state vector used by downstream signals.

The practical benefit is not that every regime gets a different strategy. The benefit is that the same strategy can scale confidence, widen protective bands, or stand down when the signal-to-noise ratio is visibly poor.

This note outlines the regime features we monitor and the failure modes that appear when those features are ignored.