Keep source data usable

Many AI workflows depend on material created for another purpose. Documents have inconsistent structure, old versions remain searchable, permissions are inherited incorrectly, and updates arrive without metadata. The resulting maintenance belongs in the operating model.

Create a source register with an owner for each collection. Record authority, scope, update frequency, access rules, retention, supported formats and the signal that marks material as current. For retrieval systems, record chunking and indexing status as derived state, not as proof that the source is valid.

Source work includes:

• adding and classifying new material;
• withdrawing or superseding old versions;
• fixing parse and extraction failures;
• reconciling conflicting documents;
• testing permissions after organisational changes;
• rebuilding indexes after a material processing change;
• checking that citations resolve to the version used;
• answering owner questions when the source itself is ambiguous.

Decide what happens when source freshness cannot be established. The workflow may block the task, warn the reviewer or limit the output. Continuing with an unknown version should be an explicit risk decision, not a silent fallback.

Measure source failures separately from model failures. Record whether the source was current, authoritative and available to the user. Otherwise a correct generation grounded in an obsolete document may be misdiagnosed as a prompt or model problem.

Budget source stewardship by collection and change rate. A stable set of approved forms has different needs from procedures updated across several teams. Include the people who understand the material, not only the engineers who operate the index.