The Python package

rnd-data (import name rnd_data) is a stdlib-only, read-only Python 3.10+ package under src/rnd_data/, licensed under MIT. Its built wheel and sdist ship only the code and schemas under src/ and LICENSE-MIT—never the data files under runs/ or store.json.

Install it

From a checkout of this repository:

python -m pip install .

The package has no runtime dependencies.

Finding the store root

Every command accepts rnd-data [--root <path>] <command>. The store root resolves in this order:

1. --root <path>
2. $RND_DATA
3. the current working directory, when it holds store.json

When none of those holds store.json, rnd-data exits with code 2 and a usage error in the studio shape (code, message and hint).

Commands

ls [--kind run|collection] [--source <prefix>] [--licence <id>]

Lists runs and collections with their kind, public and private file counts, public byte total, and the full text of every public licence in each manifest. The four early rnd-watch manifests (rnd-watch, rnd-watch--argon-gpu-stack, rnd-watch--reports and rnd-watch--runs) predate the kind field and are shown as (no kind).

rnd-data ls
rnd-data ls --kind run
rnd-data ls --source receipts
rnd-data ls --source receipts/colab
rnd-data ls --licence MIT
rnd-data ls --licence CC-BY-4.0

--source <prefix> matches by prefix: it lists every manifest whose source field equals <prefix> or begins with <prefix> (for example, receipts, receipts/colab, or rnd-watch).

How --licence matches

Licences in manifests are free text. On main there are four forms across the 871 public file entries:

• CC-BY-4.0 (557 files)
• MIT (llama.cpp, commit 620d2225) (171 files)
• MIT (138 files)
• CC-BY-4.0 (the code excerpts in context are MIT, from offrig, aspire-si and role-os) (5 files)

--licence matches a file’s leading licence identifier—the text before the first " (", or the whole value when there is none—compared exactly with no case folding:

• MIT covers the 309 files whose licence is MIT or MIT (llama.cpp, commit 620d2225).
• CC-BY-4.0 covers the 562 files whose licence begins with CC-BY-4.0.
• A licence named only inside the parenthetical note does not count: the 5 CC-BY-4.0 (the code excerpts in context are MIT, …) files do not match MIT.
• ls --licence X lists a manifest when any of its public file entries has leading identifier X. Private entries ("licence": null) and manifests with no public files never match. On main: MIT lists 69 manifests, CC-BY-4.0 lists 88, 67 hold both, and 10 have no public files.
• Each row’s licence column always shows the full text of every public licence in that manifest, never only the matched one.
• A filter that matches nothing prints an empty list and exits 0, with a stderr hint listing the leading identifiers present in the store.

show <run>

Prints one run or collection manifest in plain words: its source, ingestion date and commit, public files (with their SHA-256, byte size, tier and licence), listed private entries (private_why: identity, without any hash), content private counts by reason (private_files), and declared copies.

rnd-data show receipts--colab--rung-prestep-muse-20261019--try1-gate

where <sha256|path>

Finds every manifest name and declared copy of a public file by SHA-256 (a full 64-character digest or a hash prefix of 8+ lowercase hex characters) or by file path.

rnd-data where df19ad096ad0
rnd-data where outcome.json

where resolves blobs/<sha256[0:2]>/<sha256> only for hashes declared on public entries in public manifests, and never walks, lists, stats or opens blobs/. For a hash that no public entry declares, where answers "not declared" from the manifests alone and exits 1.

schema [--check <file>]

Prints the bundled JSON Schemas (rnd-data-run/v1 and rnd-data-store/v1), or validates a manifest or store.json file with the stdlib validator. An unknown schema value is refused.

rnd-data schema
rnd-data schema --check store.json
rnd-data schema --check runs/receipts--colab--window13l-20261018/manifest.json

Python API

rnd_data.Store(root) mirrors the CLI:

from rnd_data import Store

store = Store(".")
runs = store.runs(licence="MIT")
manifest = store.manifest("receipts--colab--window13l-20261018")
locations = store.where("outcome.json")
schemas = store.schema()
problems = store.schema(check="store.json")