Find Real-Device Regressions With MetricKit And Instruments

This article explains how to identify and resolve performance regressions that only appear on real devices, not in simulators or QA. It introduces a workflow combining MetricKit's field data collection with Instruments' local profiling. Using a SwiftUI app called Photo Loom as an example, it shows how to collect metric and diagnostic reports from real devices, group them by app state and configuration, identify regressions like scroll hitches or memory pressure, and recreate the problematic state locally with Instruments 27 to fix and verify the issue. The article also covers the new Swift-first MetricManager API in iOS 27, including async sequences and Codable reports for easy server ingestion.