Threshold Rule

A threshold rule is a decision rule that triggers an action when a quantity crosses a level: "if x ≥ T, do A; else do B." It turns a continuous or noisy signal into a discrete decision. Examples: stop when confidence exceeds 95%; launch when signups hit 1,000; escalate when latency passes 200 ms. The art is choosing T: too low and you act on noise; too high and you delay or miss. Optimal thresholds depend on costs of false positives and false negatives and on the distribution of the signal. Simple to implement and communicate; widely used in ops, product, and policy.

Define the signal and the action; choose a threshold that balances type I and type II errors (or their cost-weighted equivalents). Revisit T as costs and base rates change. Use hysteresis (different thresholds for crossing up vs down) to avoid flapping when the signal is noisy. In product: make thresholds explicit so the system is predictable and tunable.

A threshold rule triggers an action when a signal crosses a level. Choose the level using costs of false positives and false negatives; use hysteresis to stabilise when the signal is noisy. Simple and scalable for ops, product, and policy.