Reference
bnndata
A 117 GB, 564,669-file document estate for testing search, classification, indexing, and DLP tooling against the messiness of a real organization rather than a neat demo folder.
Who it is for — and why
Teams that need realistic document sprawl while still being able to tell, file by file, what was generated and what came from a public source corpus.
What was verified
- A standard-library-only generator with a fixed seed reproduces the synthetic tree
- A per-file manifest records relative path, category, extension, and byte size for every generated document
- Real public-data subsets are recorded separately from generated ones, so no synthetic file is ever presented as real
- A measured folder rollup records the file count and byte size actually on disk, independent of what the generator claims
Limits
- The repository holds the generator, manifests, and provenance — not the corpus, which is too large to distribute
- The file count is dominated by a real public-records subset; the generator's own tree is 13,838 files
- It is a test corpus, not a training set — no model has been trained on it
- The repository is private
Access and evidence
Working artifact remains private — no placeholder source or download link.