The 48-Hour Prediction Window, Explained

The 48-hour figure is a design target, not a marketing guess. It's the minimum lead time we commit to between an actionable failure alert and the predicted breakdown, so a human can schedule maintenance inside a normal working window.

It comes from the slowest failure signature we monitor. Fast signatures — a hard fault, a comms drop — give minutes, not hours. But the expensive ones (actuator wear, joint backlash, battery cell drift) develop over days, and those are the ones that ground a fleet. We calibrate the model so the alert fires early enough to order a part and book a slot.

Why 48 and not 24? Because real maintenance needs lead time: a part ships in 3–10 days, a technician slot opens, a robot comes out of service. A 24-hour alert often misses that window and becomes a reactive fix anyway. 48 hours is the floor that keeps most interventions planned.

Every alert carries a confidence score and a recommended action. The methodology page documents the full pipeline — telemetry ingestion, failure-mode modeling, and scoring.