Skip to content
rnd-datarnd-data

Overview

rnd-data is the raw data behind the studio’s research: every run’s receipts and outputs, with a manifest per run that pins each file by SHA-256. The studio is an AI and SI research studio that also makes tooling for games, and its research is open by default, so its raw data is public.

A few files have a named reason to be private: sealed or held-out keys, blind files, item text, unapproved plans, private projects, and content the studio’s identity scan flags. Those sit in the private companion, rnd-data-private, and no private file’s hash is ever published here.

What it holds

Section titled “What it holds”
Path What
runs/<id>/manifest.json One run’s files (schema rnd-data-run/v1): each file’s path, SHA-256, size, tier, licence and visibility, and every copy
runs/<id>/files/… That run’s small public files (256 KB or less), in git
store.json The stores that hold bytes outside this repository, and how a machine finds each one
src/rnd_data/ The MIT read-only Python package (rnd-data: ls, show, where, schema) and the manifest JSON Schemas
verify.sh The check CI runs, through rnd
site/ This handbook and the landing page

Files over 256 KB are stored once, by content, in a blob store, and every run that has one names it by hash. They go public as a Hugging Face dataset at the next phase; until then they’re on the studio’s backup drive.

Data, plus its read-only package under src/

Section titled “Data, plus its read-only package under src/”

The repository holds data, plus its MIT read-only Python package under src/ (rnd-data, import name rnd_data): rnd-data ls, rnd-data show, rnd-data where and rnd-data schema. See The Python package for installing and using it.

Everything that writes the data or checks the private companion is rnd’s:

  • rnd raw ingest writes runs into rnd-data, and their private files into rnd-data-private;
  • rnd raw verify re-hashes every copy against its manifest;
  • rnd where <sha256 or path> names every copy of a file, and rnd’s index reads the manifests when this checkout sits beside the rnd library, or is named by $RND_DATA.

The repository’s own scripts are verify.sh and its two workflows. The .sh and .py files under runs/ are receipts: the scripts a run used, kept as data.

Who owns what

Section titled “Who owns what”
  • R&D owns the layout and the tooling. Manifests are only ever written by rnd.
  • The Publisher owns its receipts, and reviews every ingest of them.
  • The Director rules on what is public, and gives the go before anything is deleted from the studio rig. Nothing is deleted until two verified copies exist and a restore test has passed.

Licence

Section titled “Licence”

The data is CC BY 4.0, and the read-only package under src/ and scripts (*.sh, *.py) are MIT. A third-party file keeps the licence its manifest names: the llama.cpp build is MIT (commit 620d2225).

A file’s manifest licence field is the final word for that file. Every public file’s manifest names its licence (CC-BY-4.0, MIT, or the llama.cpp build’s MIT); private entries carry "licence": null, because they aren’t published.

Where next

Section titled “Where next”