Draft zip proposal for extended CCF crawl croissants + Provenance mockup

[handecelikkanat]

@benjelloun cc. @wumpus

Sharing a draft zip file as followup to https://github.com/mlcommons/croissant/issues/961

CCF_crawl_croissants_and_provenance_mockup.zip

Zip file includes:

• 117 croissant drafts, one for each of our crawls.
• 1 mockup example for provenance citation to our crawls
  • This kind of hierarchy doesnt exist in our crawls, so we wont actually have this file in CCF, but a mockup for datasets referring to CCF.

We would like feedback especially on:

• How we are using provenance

  • eg. I havent used id's because these are not referred to in the same croissant
  • but is it valid/important to use id's to refer to other croissants?

• in how we use "distribution" with FileObjects and FileSets

  • They include a bunch of FileObjects that act as manifest files - including paths to files included in the related FileSet
    • eg., warc.paths.gz FileObject pointing to
  • And one additional FileObject example that keeps the data itself, so just a FileObject

Please let us know if anything looks awry!

Changes since https://github.com/mlcommons/croissant/issues/961:

• New FileObject added: {crawl_id}.domains-top-1000 (crawls > 2012)
• Switched to using MAJOR.MINOR.PATCH also for build version: 1.0.0+1.0.0

[wumpus]

@benjelloun can this be reviewed before 1.1 ships?

[benjelloun]

The manifest representation looks good to me.

Small suggestions for the provenance mock-up:

• the Dataset should have types: sc:Dataset and prov:Entity, since we're using prov properties on it.
• the values of prov:wasDerivedFrom should also be typed as sc:Dataset.
• I think it's fine to use URLs of datasets to refer to external datasets. If the dataset is a top-level object there, then it's not necessary to specify an id.
• Ideally the pages pointed by those URLs should contain embedded Croissant metadata.

[handecelikkanat]

Thank you @benjelloun !

I updated:

• Croissants' types as: "@type": ["sc:Dataset", "prov:Entity"]

  • Question: As far as I understand both the referencer (top-level croissant) and the referenced (second-level croissants) need to be updated as prov:Entity, so both sides of the relation should be prov:Entity, is this correct?

• and the reported types in prov:wasDerivedFrom as:

  "prov:wasDerivedFrom": [
    {
      "@type": ["sc:Dataset", "prov:Entity"],
      "@url": "https://data.commoncrawl.org/crawl-data/CC-MAIN-2025-47/index.html"
    },
   ...
]

If the dataset is a top-level object there, then it's not necessary to specify an id.

• I think this will be the case, but I will check, thank you.

Ideally the pages pointed by those URLs should contain embedded Croissant metadata.

• I think this will also be the case, yes. Adding to check.

@wumpus FYI.

[benjelloun]

• Question: As far as I understand both the referencer (top-level croissant) and the referenced (second-level croissants) need to be updated as prov:Entity, so both sides of the relation should be prov:Entity, is this correct?
• and the reported types in prov:wasDerivedFrom as:

That's correct, but we generally don't fetch the referenced croissant when validating the referencer, so it's probably okay to omit the prov:Entity type on the referenced dataset.