Naming, Chunking, and Folders

Package provenance — this chapter ships with the Ontology & Structure Engine. Provenance tag: source-ontology-project. When loaded into a new world, grep the tag to find the engine's templates and instructions.

Title tags

File names carry a tag suffix after the subject, so recall never conflates concepts that share a name. The canonical set, as shipped:

• Thornhollow-place — the place itself
• Thornhollow-lore — its history
• Thornhollow-state — its live political/cultural state

By extension the same pattern covers anything that could collide with a same-named sibling: -person, -rule, -spell, -event, -memory, -clan. The convention is soft: apply the tag where it adds recall value — places, named things, anything two searches could both return. A tagless name is fine when the subject is unambiguous.

Why it matters: embedding retrieval matches on meaning, and "Thornhollow" means three different things in this knowledge base. The tag is cheap disambiguation that costs nothing at recall time.

One file, one fact cluster

A file is one retrievable idea. Split a file when:

• its body has grown past a single cluster — two paragraphs that could be recalled separately belong in two files;
• its keyFacts stop sharing one subject;
• a reader looking for one of its facts would be annoyed to retrieve the other.

When you split, keep the pieces linked: the old file's relatedLore points at the new ones. Deleting or merging clusters is the opposite operation — do it only when two files genuinely never get recalled apart.

Folder nesting

Organize by ontology domain, deeply:

/Lore/
  Geography/
  History/
  Culture/
  Magic/
  Rules/

Nest subfolders for regions or civilizations when content grows (/Lore/Geography/The Marches/Mistveil/), and keep the three filters in mind: folder is the coarse filter, tags the fine filter, references the graph. A flat pile of files defeats retrieval no matter how good the tags are — but a folder should always mean something, never just be a drawer.