Find It in Production
Decide what you would build from the brief alone. Everything below it is available, but the exercise stops working if you open it first.
p99 latency on the main API tripled over a week. p50 is unchanged. Error rate is flat. There were eleven deploys, a database version upgrade, and a large new customer onboarded. Find the cause.
Rolling back the database upgrade, since the timeline fits and a version change is a satisfying explanation for a plan regression. It is a real suspect and it is also the most expensive move available — hours of work, a maintenance window, and if the tail does not improve you have learned only that this one hypothesis was wrong, while the segmentation that would have pointed at the new tenant in minutes is still not done. Cheap discriminating evidence first; expensive reverts last.