Datasets

There are no accepted standalone datasets in the current reference asset set. Notebook outputs and generated figures should not silently become datasets merely because they are saved to disk.

A dataset is accepted only when its provenance, license, schema, checksum, and reproduction path are clear.

Dataset Types

Future datasets may include:

benchmark reference values,
exported convergence tables,
generated spectra for standard models,
public-domain historical measurement data,
package-comparison outputs,
figure source data,
notebook execution reports.

Each type needs a different review standard. A small table of exact oscillator energies is not the same kind of source as an external experimental dataset.

Admission Checklist

A dataset should record:

path,
creator or source,
license,
schema or column definitions,
units and conventions,
generation script or notebook,
checksum,
date generated or accessed,
page using the data,
review status.

Do not commit large generated datasets without a plan for maintenance and reuse.

External Data

External data requires extra caution:

cite the original source,
record the access date when data can change,
keep the original data separate from processed data,
document preprocessing steps,
preserve units and metadata,
respect license and attribution requirements.

Historical datasets should distinguish the original measurement from a modern reconstruction or digitization.

Generated Data

Generated data should be reproducible from tracked source. If a plot uses sampled values from an analytic formula, the formula and sampling grid should be recoverable. If a numerical notebook generates the data, the notebook should carry environment and benchmark status.

Cross-Links

References

The Turing Way Community, The Turing Way: A Handbook for Reproducible, Ethical and Collaborative Data Science.
Creative Commons, CC Licenses.
SPDX, Handling License Info.