Corpus Authorship

Who wrote the body of text being searched. In retrieval, someone else did: documents, tickets, a wiki. In a memory layer, the system writes its own corpus as a byproduct of interacting.

An authored corpus is small, mutable, contradictory, and about one person. A given corpus is large, static, internally consistent, and about a domain.

Why it matters in a memory layer

Nearly every design difference follows from this one distinction. A given corpus is chunked because it is too big; an authored corpus is extracted because raw turns are the wrong unit. A given corpus is versioned by re-ingestion; an authored corpus mutates in place and needs an audit trail. Getting this backwards produces a RAG pipeline with a user_id column and calling it memory.

Connections

Taught in: Memory Is Not RAG