This is the part skipped at launch and the part that decides whether the project survives. A taxonomy with no owner and no change process degrades predictably: duplicates appear first, then catch-all terms like “other”, then people stop tagging altogether.
What we settle
Who owns the vocabulary and holds the veto. What a new-term proposal looks like and how fast it gets answered. What happens to a retired term — deleting breaks links, keeping it adds clutter. How often review happens and against which checklist.
Term lifecycle in production
A governed vocabulary treats every structural edit like a software release rather than a casual spreadsheet row:
- Input signal. Candidate terms surface from three places: zero-result search queries, monitoring of the residual bucket (“other” crossing a 3–5% quarterly threshold), or direct requests from taggers.
- Quarantine and canonicalisation. Before minting a new node, we check whether the proposed label is a lexical variant of an existing concept (in which case it becomes an
altLabelentry synonym) or an attribute belonging to another facet. - Publication gate. A new leaf enters production only when it has a clear parent, a yes/no boundary rule separating it from its neighbours, and enough real corpus items to justify its existence.
- Deprecation without broken links. Retired terms are never hard-deleted from the database: they are locked against new tagging while keeping their stable ID and a mapping rule to their successor, preserving historical analytics and deep links.
Two engagement models
- Governance setup (one-off): designing the ownership roles, change-request workflow, decision log and review checklists, closed with a practical training session for editors and vocabulary stewards.
- Periodic stewardship (quarterly retainer): ongoing care for a live vocabulary — quarterly audits of new tagging and the “other” bucket, adjudication of contested boundary cases, and versioned vocabulary releases (
v1 → v2) with a complete changelog.