How would you measure P95 latency by geographic region?

This tests your ability to design a full metrics pipeline, from instrumentation to query. It assesses your understanding of metric types (histograms), metric vs. log trade-offs, and high-cardinality dangers. A great answer clarifies the region source (GeoIP), instruments the API to record latency as a histogram with a `region` label, and provides a pseudo-query using `histogram_quantile`. A common mistake is suggesting only logs, which are expensive, or proposing a metric with a user-ID label, which causes a cardinality explosion.