data

Data

Schemas, catalog exports, and reproducibility contract.

Public catalog

The generated catalog JSON is the site’s machine-readable index. It contains every experiment, model, and run shown in the UI.

Canonical source

Git remains the source of truth. Records live under data/experiments, data/models, and data/runs. Artifacts live under data/assets in v0.1 and are duplicated into the Hugo build by the deterministic catalog compiler.

Schema contract

The repository publishes Draft 2020-12 schemas for experiments, models, and runs. CI rejects unknown fields, malformed identifiers, broken references, missing artifacts, hash drift, MIME mismatch, secret-like query keys, duplicate IDs, duplicate asset references, and orphan assets.

Reproduce locally

python -m venv .venv
.venv/bin/pip install -e '.[dev]'
.venv/bin/python scripts/validate.py
.venv/bin/python scripts/build_catalog.py
hugo --source site --minify --panicOnWarning

The full source and issue board are available in the public GitLab repository.