A data manifest is a file delivered with a batch of data that describes the batch: the files it holds, the period and sources it covers, and the terms under which it is supplied. A receiver reads it to check that a delivery is complete and to record what was loaded.
What a manifest is used for
A manifest turns a delivery into something you can test. A sound load handles it in this order:
- Keep the manifest with the files it describes.
- Check that every file it lists is present, and that sizes, row counts or checksums match where it states them.
- Compare the period with the one you asked for.
- Compare label and schema versions with those of the previous delivery. A new version name is a decision to make, not a surprise.
- Store the manifest with the load, as the record of what you loaded and under which licence.
The fields differ between providers, so read a sample before you write the checks. An illustrative layout, as JSON, for a delivery of hiring tables:
{
"delivered_at": "2026-10-05T06:00:00Z",
"period": {"from": "2026-09-28", "to": "2026-10-04"},
"sources": ["company job boards"],
"label_versions": ["jobs-v2"],
"licence": "internal use",
"files": [{"name": "company_hiring_daily.jsonl", "rows": 12000, "sha256": "..."}]
}Each key becomes a check: the period against what you asked for, the versions against the last load, the file list and counts against what arrived, and the licence against your agreement.
Manifest, dictionary and changelog
A manifest describes one delivery. A data dictionary describes the structure of the dataset and changes only when the schema does. A changelog describes what changed between versions. You need all three: the dictionary to read the columns, the manifest to know what you hold, and the changelog to know why two deliveries differ.
In Fokals data
A packaged export of files for a period comes with a manifest naming sources, period, label versions and licence. The export endpoints of the API return the same rows page by page, for the period you name in the request. The layout above is illustrative; a Fokals manifest names sources, period, label versions and licence. The label versions are the part to watch. Labels and scores are produced under named, frozen versions, one for postings and one for websites, and a breaking change ships as a new version with at least 90 days' notice. When the version in a manifest differs from the last one, read the methodology for what changed before you load.
The licence line gives a team that passes data on a record of the licence each delivery was made under. For the exports themselves, see the delivery page.
Related terms
- Bulk export: the delivery of data as files, which a manifest accompanies.
- Label version: the frozen name under which labels are produced.
- Redistribution licence: the licence whose terms a manifest helps you evidence.
- Data provenance: the sources behind each record.
Frequently asked questions
What is in a data manifest?
A manifest lists what a delivery holds and where it came from. Common contents are file names, sizes, row counts or checksums, the period covered, the sources, schema or label versions and the licence. The fields vary by provider, so read a sample first. A Fokals export manifest names sources, period, label versions and licence.
Is a data manifest the same as a data dictionary?
No. A dictionary describes the structure of a dataset and stays the same across deliveries until the schema changes. A manifest describes one delivery: its files, its period and the versions and terms that applied to it. You use the dictionary to interpret the columns and the manifest to show what you loaded and when.
Why keep a manifest after loading the data?
A stored manifest is your record of the period, sources, versions and licence a table came from. It answers an auditor's question months later, explains why two loads differ when a label version changed, and shows a redistributor what each customer was sent. Store it in a table beside the data, with the time of the load.
The queries and code on this page are examples to adapt. Test them in your own environment before you rely on them.