Skip to content

Datasets

A dataset holds metadata entities organized according to a profile specification. Datasets are the main objects you work with in the hub; the Datasets page lists every dataset you own or that has been shared with you.

The dataset list

Open Datasets in the header. The list holds every dataset you own and every one shared with you, in two views you switch between with Cards and Table above the list; the hub remembers which you used last.

Cards show each dataset with its profile, version, last-updated date, and how many entities it holds; hovering over the count shows the number per entity type. A dataset shared with you through a collaboration carries the collaboration's name.

Table shows the same datasets one per row, with the columns Name, Profile, Version, Entities, Updated and Access, and is where you find one among many:

  • Search matches the name, the description and the profile as you type; the list narrows without reloading the page.
  • Profile narrows the list to one profile; the choice lists the profiles your datasets use.
  • Access narrows the list to datasets you own, datasets shared with you by a person, or datasets reached through a collaboration.
  • Click a column heading to sort by it; click again to reverse. The list arrives sorted by last update, newest first.

The filters live in the page address, so a filtered table can be bookmarked or sent to a colleague, who sees it narrowed the same way over their own datasets. Click a row to open the editor.

Creating a dataset

  1. Click + New Dataset.
  2. On the Standards tab, select a profile and version. The newest version of each profile carries a latest badge. The supported profiles are listed in the profiles reference.
  3. Click Create, or With Example to start from the profile's worked example.
  4. Name the dataset in the dialog that opens and confirm. The name is asked after the standard and version, as the local metaseed app asks it; a file import suggests the file's name, an example the profile's, and a repository import takes the name from the record and asks nothing.

A dataset name is unique within your account, and a deleted dataset keeps its name. A name you already use, or one held by a dataset you deleted, is refused and nothing is created; the form stays open and says why, so you can choose another name. The same applies when you import a file or a repository record. When creating a dataset fails for another reason, such as an accession that resolves to nothing, the form says that instead.

To start from a file instead of an empty profile, use the Import File tab. See Importing and exporting.

The dataset editor

The editor has three areas:

Area Contents
Left sidebar + <Entity> buttons for entity types not yet added, the dataset actions, and two tabs: Entities (the entity tree) and Sharing
Center The entity overview with History and Comments tabs when no entity is selected; otherwise the form for the selected entity

From the sidebar you can also Validate the dataset, open the Graph view, Import into the dataset and Export it. Delete stands alone at the foot of the sidebar, apart from the routine actions.

Specification drift

Every save records the content hash of the specification the dataset was authored against. When you validate, the hub compares that stamp with the specification's current hash and reports specification drift if they differ — the specification changed after this dataset was written, so entities that were complete when you saved them may no longer be.

Drift is reported, not enforced: the dataset still opens, still edits, and still validates against the current specification. The report tells you why a dataset you had finished suddenly has issues.

Datasets saved before the hub started recording the stamp have none. Their provenance is unknown rather than unchanged, so no drift is reported for them; the stamp is recorded the next time the dataset is saved.

See:

Deleting a dataset

Click Delete (the red button at the foot of the dataset sidebar) and confirm. A deleted dataset is removed from your list and excluded from all queries. Deletion is a soft delete: see Versions and history for what this means.

FAIR exposure

A dataset page is also its catalog record. The page embeds the dataset's DCAT description as JSON-LD, the dataset URL answers content negotiation (Accept: application/ld+json or text/turtle returns the record itself), and every response carries a Link: rel="describedby" header — the three signals FAIR harvesters such as F-UJI read. Reachability follows the deployment's authentication: harvesting needs whatever visibility the operator grants the dataset URL.

An opt-in FAIRness regression check runs F-UJI against a reachable dataset URL (tests/test_fuji_fairness.py; set FUJI_URL and FUJI_TARGET). It asserts the FsF score does not regress below a baseline, which is how a change that quietly removes a harvestability signal gets noticed.