Measuring Black-Hat Signals Without Burning Domains

I isolate experiments and track three metrics only time-to-peak, volatility window, and recovery cost then use the signals to design safe tests or white-hat equivalents. Who else treats risky tests as research data rather than production tactics?