[falcolas]

For me/us, it's a combination of infrastructure-as-code and metrics reporting/logs. Most of our boxes are swapped out on a weekly or more frequent basis, so the only accurate picture of what's running right this moment is the graphs built by the metrics collection tools. The only accurate picture of what's running on those boxes is the code which built the infrastructure.

There are a couple of exceptions, but those are actively being brought under the above model (mostly because they are effectively invisible, and the existing documentation for them is... incomplete).

Any documentation outside of that is stale in a few hours, and obsolete in a week.