3. Catalogue in YAML, not SQLite

Context

The catalogue holds several hundred procedure descriptions and is read once per process. The original repo seed proposed data/catalogue.yaml compiled into SQLite.

Decision

The catalogue is a single YAML file, parsed and cached at first use. No SQLite, no build step, no compiled artifact.

Consequences

The catalogue stays diffable, reviewable in a pull request, and directly publishable as CC BY open data — which is the point of keeping it separate from the code at all. A few hundred rows parse in milliseconds, so the database would buy nothing but a second source of truth. Revisit at more than a thousand entries, or when the documentation gallery needs to query rather than iterate.