Rules
Velocity rules that age badly
8 min read · Schema Designlab Journal
Velocity rules feel precise because they are numeric. That precision is often historical. A threshold chosen during a spike in mule activity can quietly punish ordinary behaviour a year later when product, geography, or payment rails have changed.
Symptoms of an aged rule
Analysts stop opening the alert because “it is usually fine.” Product files tickets about blocked power users. Nobody owns a review date. The fraud signal audit app still shows a green “active” badge as if the rule were freshly justified.
Age itself is not the problem. Missing review cadence is. We teach teams to attach a next-review date to every velocity family and to treat missed dates as findings, not reminders.
A lighter rewrite pattern
Instead of deleting a rule in panic, write the claim it still intends to prove. Compare current false-positive samples against that claim. If the samples no longer illustrate the claim, narrow the window, add a compensating attribute, or move the alert to a manual-only path with an explicit note that auto-close is forbidden.
Our Signal Clinic format exists for teams who need a half-day on one stubborn velocity family rather than a full cohort.