An endpoint's average latency is a steady 50 ms. Why might users still complain that it is slow?

Options

• ✓ The average hides the tail: 2% of requests taking 2 s barely moves it Correct
• ✗ Averages are always measured on the server, not the client
• ✗ 50 ms is too slow for any endpoint
• ✗ The average counts failed requests twice

Why

98 requests at 10 ms and 2 at 2 s average about 50 ms, yet one in fifty users waits 2 seconds, and a heavy user making dozens of calls hits that often. Track p50, p95 and p99 (or p99.9 for large fan-outs) instead.

Review this in your daily deck → All cards in Observability