Grafana dashboard notes: fewer panels, better questions

My first home-lab dashboard had too many panels. Every sensor had a gauge, a current-value card, a graph, and a status indicator. It looked active but answered simple questions slowly. After a few months I rebuilt it around the questions I actually asked when something looked wrong.

Question one: did the room change, or did the sensor change?

The main climate panel now plots temperature from three nearby sensors on one graph. A single line can show a trend, but three lines show agreement. If all three move together, the room probably changed. If one jumps alone, I inspect the device or its placement. I keep the raw values; no smoothing is applied to this panel.

The next panel shows humidity on the same time range. Temperature and humidity are separate because their useful vertical scales are different. Combining them on two axes made the graph compact but much harder to read.

Question two: is the data fresh?

A small row at the top shows sample age rather than a generic green state. The query is simply the latest metrion_sample_age_seconds value for each sensor. Thresholds are 120 seconds for warning and 300 seconds for stale. That makes a slow node visibly different from a completely silent one.

Question three: what changed before the event?

I use one 24-hour overview and a default six-hour detailed range. Fan duty, enclosure temperature, and supply power share aligned time axes in separate panels. This is much more useful than three dashboards with unrelated zoom levels. When the enclosure temperature rises, I can immediately see whether fan duty increased and whether electrical load changed at the same time.

I removed decorative gauges entirely. A gauge is attractive for a fixed acceptable range, but most of these values are meaningful because of change over time. I also stopped using more than three threshold colors. Normal, attention, and stale are enough for this setup.