Why does logging quality usually fall apart as systems grow?

One pattern I keep seeing is that logging starts out useful, then gradually becomes inconsistent, noisy, expensive, and harder to trust. Field names drift, context goes missing, dashboards get polluted, and sometimes sensitive data ends up in places it never should have reached.

The deeper problem seems to be that most teams try to fix logging after the data has already entered downstream tools. By then, the cost, risk, and cleanup burden are already there.

I’m curious how other teams handle this. What breaks first in practice: naming consistency, missing required context, sensitive fields in logs, alert noise, or ingestion cost? I’m building in this area and want to learn where current approaches still fall short.