Asset Card and Observed Lineage

Before trusting a dataset, its users need three answers: what it is, where it comes from and who is accountable for it. The Asset Card gathers the pieces of information that answer those questions in a single view — and the Lineage tab shows, from what actually ran, who reads and who writes each asset.

Where the card appears

The same card opens in three places:

• Knowledge Catalog › Discovery — click a dataset card. The detail panel shows the full card and, below it, the column list.
• Knowledge Catalog › Knowledge — in the semantic search, click a result: the detail shows the card and the LLM classification.
• Data Governance — after selecting a dataset, the card appears below the current governance.

The dataset is identified by the natural key shown in grey under the name (for example neo4j:FEBRABAN:Empresa or opensearch:datta_bcb). It is the same key across all screens, the API and the lineage graph.

The sections

Each section loads independently: if one is unavailable, the others still appear.

Columns: the data dictionary

The Columns section answers "what does this column mean?" — the question the asset description cannot reach. Click Describe on the column row, write up to 300 characters and save; the text becomes visible to everyone, with who wrote it and when.

A column that has not been described yet stays on the list, with its type and the invitation to describe it: no asset loses the section for lacking the field. The counter next to the title (3/28 described) shows how much of the dictionary is written, and the filter finds a column by name on wide assets.

The description is curated by people and only changes when someone edits it. Nulls, uniqueness and distribution of the same column are measured and live in the Quality section, rewritten on every profiling run.

Certification

The ✓ Certified badge tells the analyst "this is the reference asset". It appears on the Discovery card, in the Knowledge search results and on the node in the Lineage tab. Deprecated warns that the asset should no longer be used — use the Reason field to say what to use instead. In the Knowledge search, certified assets come first, and the Certified only box filters out the rest.

Certifying requires the DATASET_CERTIFY permission (administrator and advanced user). Without it the button answers "You do not have permission for this action".

Documentation and search

What you write in the Documentation section is stored in a dedicated index and vectorised with the platform's active embedding model. The Knowledge search then finds the dataset by the documentation text — the result is marked via documentation when only the documentation matched. If the embedding service is down, the text is stored anyway and the vector waits for the next save.

Tags in search

In the Knowledge tab, the Filter by tag field restricts the search to datasets carrying that tag. Tags are normalised (lower case, hyphen instead of space): Personal Data and personal-data are the same tag.

Observed lineage

The Knowledge Catalog Lineage tab shows the graph of datasets and processes in three layers:

• Observed (default) — what actually read and wrote data: every run of a Datta Extract package, every document ingestion and every consumption by chat or report. Arrow thickness grows with the number of runs.
• Architecture — the declared topology of the platform services (only exists when the administrator enables the architecture bootstrap).
• All — both together.

Select a node to see, in the right-hand panel, how many times it ran, the last run, and what it comes from and what it feeds. On a dataset, the Open in catalog button leads to the Asset Card. The context filter follows the active workspace.

If the tab shows "No observed run", no Extract package has run in that context yet: run a package and come back — the lineage appears as soon as the run finishes.

Permissions

Every change is kept in the audit trail with who did it and when.