Trace replay lets you walk through the exact sequence of model calls, tool results, retrieval hits, retries, and approvals from a previous run. It is how a vague complaint becomes a reproducible bug: you can see where the context went wrong, swap a prompt, and compare the new run against the old one.
ELI5
Think of Trace Replay as the checklist that keeps the app open for real people. Re-running or inspecting a saved agent trace step by step to understand, debug, or reproduce behavior.
In practice
Use it when a change has to survive deploys, users, incidents, analytics, or billing reality. In practice, define the owner, input, output, and failure mode before you rely on it.