The dashboard is not the reality

We have dashboards. That does not mean we can see.

It has become obvious during the reliability work. The systems are instrumented in a lot of places: crashes, latency, funnel drops, complaint categories. But measurement can manufacture a false sense of sight, and the gap shows up exactly where the instruments don’t reach.

The dashboard shows what we thought to measure. The user feels everything else.

A parent does not know which backend owned a failure. They do not know it was stale policy, or a sync that did not happen, or a surface that forgot to handle a supervised account. They know the thing they trusted did not behave the way they expected. That gap is humbling, and it tells me reliability is not only driving down known error rates. It is improving our ability to find out what reality feels like outside the system.

So I have started to think a mature product needs two kinds of observability. One for the machine: is the system functioning. One for the promise: is the person getting the experience they believed we offered. The second is harder. It means mapping journeys across team boundaries, connecting complaints to system behavior, and asking engineers to care about signals that do not look like clean logs.

And it asks something the first kind never does. A green dashboard is comfortable. Believing it while users tell a different story is a choice to not look. The real reliability question underneath all the metrics is whether the organization can tolerate seeing itself clearly. Most of the work was getting us to want to.