Product stance: DocuMind is Wikipedia-primary for scale (library=public + offline bulk index). This directory is a small synthetic / curated bundle for optional papers demos—not a resume-grade “millions of documents” knowledge base.
| Metric | Count |
|---|---|
Total *.txt files |
463 |
Synthetic sample_corpus_p7_*.txt |
400 |
Other named / hand-authored .txt |
63 |
Chunk row count is not stored in git: it is whatever Chroma holds after SEED_SAMPLE_DOCS=true startup indexing (CHUNK_SIZE, CHUNK_OVERLAP, and text length). Order of magnitude for the full bundle: low thousands to ~10k+ chunks (estimate only).
Contrast: English Wikipedia is on the order of millions of articles (CC BY-SA). That is the credible production RAG scale story; use scripts/stream_wikipedia_to_txt.py + scripts/bulk_index_public.py into CHROMA_COLLECTION_PUBLIC.
CI and unit tests use fake corpora (e.g. tests/ranking_fake_embedding.py — six deterministic chunks), not this directory.