I've now watched three companies build the same thing. A central layer where you define "active user" or "revenue" exactly once, so every dashboard agrees. dbt metrics, LookML, a homegrown YAML registry, the technology varies. Every time it's sold as the end of the "your number doesn't match my number" meeting. And every time, within about a year, the definitions have drifted, half the dashboards bypass the layer with raw SQL, and we're back in the meeting.
My question for people who've kept one alive longer: is this fundamentally a tooling problem or a people problem? Because I keep concluding it's people, and I want to be wrong, because people problems don't have a roadmap.
My read is that a metric definition isn't a technical artifact, it's a political settlement, and settlements expire. The moment "active user" produces a number some VP doesn't like, the pressure isn't to fix the data, it's to quietly define a new "engaged user" next to it that says what they need. The layer doesn't rot because the tool is bad. It rots because nobody with authority is willing to be the one who freezes the definition and eats the unflattering number.
So: has anyone actually kept one of these honest past a year, and if so, was it better tooling or someone senior willing to enforce it?