At some point I crossed the line from monitoring my homelab to collecting dashboards. I counted them last night: eleven. Node exporter overview, a per-host one, a ZFS one, a Proxmox one, two for Pi-hole that I do not remember making, a "network" one that was mostly empty, and a few more I clicked past quickly out of embarrassment.
The honest test is this. When the NAS started throwing read errors on Sunday, did any of those eleven tell me? No. I found out because Plex stopped playing and I went looking. The dashboards were all green because none of them were watching the thing that actually mattered, they were watching whatever Prometheus happened to scrape and Grafana happened to render nicely.
So I deleted nine of them. What I kept is one overview with the handful of signals I'd actually be paged on if this were a job: disk health, pool capacity, a couple of host-up checks, and free memory on the box that runs everything. Then I wired those few to alerts that ping my phone, because a graph nobody is looking at is just a screensaver.
The lesson, again, because I clearly need it written down: a dashboard you glance at is not monitoring, it is decoration. Monitoring is the thing that wakes you up. Eleven pretty panels did not wake me up. One ugly alert rule would have.