App startup is slow for some users. You know how long it takes, but how do you figure out where that time is actually going? This talk is about using traces to debug performance problems in production Android apps. We'll take that slow launch and try to account for where the time went. We'll walk through the resulting trace and break the launch down into the work happening across the app. Along the way, we'll look at how to design traces that are actually useful for debugging: what should become a span, what belongs in an attribute, how different sampling strategies let us collect useful traces at scale, and how distributed tracing can follow the time across network requests and backend services. We'll also look at where traces fit with the other tools available for debugging performance. Metrics can tell us that something is slow, traces can narrow down where the time is going, and when that still isn't enough, profiles and system traces captured with Android's ProfilingManager APIs can help us go deeper. You'll leave knowing how to design useful traces, investigate slow operations with them and recognize when you need to go deeper with profiling.