Schema
Records that hold their shape.
A knowledge record for LLM use needs more than a definition. It needs fields that explain identity, context, boundaries, and reuse conditions. The schema is the contract between writers, retrieval systems, and answer surfaces.
Canonical name
The stable label that should survive prompts, exports, and retrieval indexes.
Alias set
Accepted variants, discouraged variants, and names that look similar but mean something else.
Boundary note
A plain-language statement of what the record excludes.
Evidence hint
The kind of source, date, or observation that supports the record.
Answer summary
A concise form that can be reused in answer engines without flattening nuance.
The schema is not decoration.
When fields are optional only in theory, records become inconsistent. LLM Wikibase treats naming, boundary, and evidence fields as operational requirements because they decide whether future retrieval can distinguish a useful record from a vague note.
Short summaries still need structure.
A compact answer field should be paired with the longer record that produced it. That pairing lets a system cite a clear sentence while preserving the context needed for review, correction, and future expansion.