pub fn embedding_text(memory: &CanonicalMemory) -> StringExpand description
The text to embed for a record.
The statement, followed by the record’s own frontmatter written as prose. This is the measured-best input to a dense retriever for this corpus — see the module docs for the comparison it won.
Use this in a SemanticFallback implementation
so that what is indexed matches what was measured:
for record in records {
let text = embedding_text(record);
// embed(&text) and store the vector against record.id
}