The catalog

The largest source of duplicated processing in an enterprise is not technical; it is that nobody knows what already exists. Team B builds an extraction pipeline over a document set because they did not know Team A had already done it, registered nowhere, discoverable by nobody.

The catalog is the answer. Every knowledge product registers there with its contract: scope, owner, representations, metadata, freshness, quality, access requirements, endpoint. A consumer — a person designing a new use case, or an agent choosing a tool — searches the catalog before building anything.

For the catalog to work, registration has to be the path of least resistance. If the platform's ingestion service registers the product automatically as a side effect of onboarding, the catalog stays complete. If registration is a separate manual step, it decays within a quarter.

The catalog also carries the cross-domain view that no single domain can: which products overlap in scope, where source-of-truth ranks conflict, which products are consumed by whom, and where the same source system feeds more than one product. Those are the governance questions, and the catalog is where the evidence to answer them accumulates.