Learn / Feature intuition
Answer page / feature intuition
Topic cluster / Regime detection and contextThe best context layers usually describe conditions that change the payoff of the core signal, such as volatility stress, liquidity quality, crowding, trend persistence, or event timing. They should explain when the base edge strengthens or weakens, not just add another correlated feature for its own sake.
What to remember
A context layer is worth testing when it describes why the base strategy works better or worse in certain environments. That is a different standard from simply asking whether the extra feature correlates with returns somewhere in history.
For example, a short-horizon mean-reversion sleeve often cares about volatility spikes, book quality, and crowding stress long before it cares about a slower macro narrative.
The useful categories are usually practical rather than mystical. They describe conditions that affect execution, edge persistence, or capital competition.
Do not add a context layer just because the backtest improved after another feature was appended. If you cannot explain what decision the layer changes, it is probably just another version of the signal wearing a different name.
Alphora's public pages already hint at the right pattern: realized volatility as context, premium and funding as different views of crowding, and simulator tools that make it easier to see how a new layer changes the whole portfolio rather than one isolated equity curve.