The catalogue nobody browses

Nearly every organisation has bought or built a data catalogue. Very few people open one. The usual diagnosis is that the metadata is incomplete, so the next project is a documentation drive, which produces a burst of descriptions that decay over the following year, and the catalogue is no more used than before.

The diagnosis is wrong

The problem is not coverage. It is that browsing is not a thing people do. An engineer with a question does not go to a catalogue and explore; they search for a name they half-remember, or they ask a colleague, or they read the SQL. A catalogue competes with those three, and against a colleague it loses on latency and against the SQL it loses on trust.

What a catalogue is good for

• Answering a specific question asked in words, rather than presenting a tree to walk.
• Being the substrate other things are computed from: lineage, cost attribution, dead-object analysis, drift.
• Telling you what exists that you did not know existed — which is a report it should push, not a page you should visit.

Read that way, the catalogue stops being a product people use and becomes an index other features depend on. The measure of it is not visits. It is whether the questions built on top of it can be answered, and how often the answers are right.

Judge a catalogue by what it makes computable, not by who opens it.