Every request generates a log line, and large sites produce billions of rows a day, so teams often keep only 1% or 10% of the logs. Sampling leaves blind spots: errors at a single edge node can be dropped entirely, and the dashboard looks healthy while real users stay stuck.