Datasets

If your research project produces a dataset, you can indicate it as such to make it easy for others to reuse in their own project. These are listed in the datasets section of the project's calkit.yaml file.

A dataset is identified by its path in the project repo, and this path can be a folder. For example:

# In calkit.yaml
datasets:
  - path: data/raw-data.csv
    title: Raw data
    description: This is the raw data.

The purpose of the datasets section is to list those that might be useful to others in their project and to note which were not produced as part of yours, i.e., they were imported from elsewhere.

Importing or reusing a dataset from another project

A dataset can be imported with the CLI like:

calkit import dataset {owner_name}/{project_name}/{path} {local_path}

If this dataset is tracked with DVC, a new DVC remote will be created to pull it into your project. For datasets on the Calkit hub, this means the data will not be duplicated there.

Declaring an imported dataset

If you imported a dataset yourself from a web URL, DOI, or a Git repo, say so like:

datasets:
  # From a website
  - path: data/from-elsewhere.csv
    imported_from:
      url: https://website.org/data/something.csv
      date: 2021-01-01 # Optional
  # From a DOI, e.g., Figshare or Zenodo
  - path: data/from-archive.csv
    imported_from:
      doi: 10.5281/zenodo.1234567
  # From a Git repo
  - path: data/from-git.csv
    imported_from:
      git:
        repo_url: https://github.com/someone/something.git
        path: data/their-dataset-name.csv
        rev: 4031e49efbea3be3b6b10e66f30d7cff6dfc60cc

Declaring a dataset you collected yourself

If you manually collected a dataset, i.e., it is a primary artifact, that should be declared explicitly with created_by, naming whoever collected or measured it, the same key a hand-drawn figure or a photo uses:

datasets:
  - path: data/raw.csv
    created_by:
      email: me@myorg.edu
      orcid: 0000-0002-1825-0097 # Optional

created_by can be a list, since data is usually collected by more than one person, and each entry carries its own identifiers:

datasets:
  - path: data/raw.csv
    created_by:
      - email: me@myorg.edu
        orcid: 0000-0002-1825-0097
      - orcid: 0000-0001-5109-3700
      - email: acolleague@elsewhere.edu
        name: A Colleague

Each person needs an email or an ORCID, or both. A name on its own doesn't say which of the several people with that name this is, so credit rests on something resolvable. An ORCID is the better of the two, since it identifies someone beyond this one project, and it's stored as the full https://orcid.org/... URL whether you write it that way or not. That's the form CITATION.cff uses too, so the same value can be copied between the two. Its last character is a check digit, so a mistyped one is refused rather than recorded.

A DOI can be written bare, as doi:10.5281/zenodo.1234567, or as the https://doi.org/... URL; it's stored bare. A created_by and an imported_from on the same dataset is an error, since data is either something you produced or something you got.

Entering data by hand

Some data is typed in rather than downloaded: readings off an instrument, a tally, a table copied out of a paper. On the hub, the dataset dialog's "I'll type it in now" option opens a small grid for exactly this. Paste a block from a spreadsheet to fill many cells at once. Saving writes a CSV at the path you choose, tracks it with DVC, and records you as its collector, so it gets the same provenance as any other primary data:

datasets:
  - path: data/readings.csv
    title: Bench readings
    description: Voltage and current at each load step, 2026-08-20
    created_by:
      - email: me@myorg.edu

Treat it the way you would a lab notebook page. If the grid was a stand-in while real data was on its way, replace the file rather than leaving the placeholder declared as collected data.

For attributing files that aren't datasets, and for disclosing generative AI, see provenance.