The operations review combines a nested event feed with a slow reporting job. The data questions need exact rows. The performance questions need evidence about the workload rather than an assumed speedup from a command name.
Reuse JSON paths, grouped collections and layout concepts. Keep the parent identifier when you extract event details, and remember that collecting or exploding values changes the shape of a result. For runtime problems, compare stage metrics and the input processed by each task. A large shuffle, a single hot key and memory spilling can require different responses. The local simulator deliberately does not report production performance gains from compaction or caching. Your answer should identify what you would measure in the real environment and why that evidence fits the stated symptom.