Governing source data inline

Business entities and attributes are where governance is meant to live: a definition, a steward, a classification, all in one place that source tables and columns read from once they are mapped to it. Mapping takes modeling work, though, and sometimes you notice a sensitive column before the attribute that should own its meaning exists yet. Inline governance lets you write a definition and set a classification directly on an unmapped source column, so nothing has to wait for modeling to catch up. The moment that column is mapped, the business attribute becomes the authority and your inline notes step aside.

Inline governance versus mapped governance

A source column is governed one of two ways, depending on whether it is mapped:

Unmapped: the column carries its own inline governance, a definition and a classification stored directly on the row. You write and edit these yourself; nothing propagates from anywhere else.
Mapped: the column is linked to a business attribute, through Map and match or a direct pick. The mapped attribute becomes authoritative for the definition, exactly as described in Data governance overview. Classification is the exception: you can still set it locally on the mapped column, covered below.

Once a column is mapped, its inline editors step aside for everything except classification. A mapped column’s governance panel shows a short read-only note pointing at the business attribute instead of a definition box, and its classification renders read-only too, labeled with where it comes from. To edit the definition itself, open the attribute it is mapped to, from the modeling canvas or the entity’s Attributes tab, and govern it there.

Turning governance requirements on

Inline governance and the panels described below are opt-in. Nothing changes for your organization until an administrator turns requirements on.

Requirements live on the Data Governance page, under Settings → Data Governance (/admin/settings?section=data-governance). Like Conventions, it is a single organization-wide setting rather than a project or workspace one, and only an organization administrator can change it.

The page has three parts. Two of them decide what is tracked:

Enable governance requirements: a single switch. Off (the default) hides every governance panel, badge, and filter described in this guide, everywhere. Turn it on to start tracking.
The requirements matrix: a row per asset type, business entities, business attributes, source tables, and source columns, with a control per governable property. Each property is set to one of:
- Off: the property is not tracked and does not appear in the governance panel for that asset type.
- Recommended: the property appears in the panel as worth filling in, but an empty value never counts as incomplete.
- Required: the property appears in the panel and counts toward completeness. An empty required property is what makes an asset show as incomplete, on the badges and filter covered below.

The properties available differ by asset type:

Asset type	Properties
Business entities	Definition, Steward, Owner, Category, Classification
Business attributes	Definition, Classification
Source tables	Mapped or defined, Classification
Source columns	Mapped or defined, Classification

Mapped or defined ties back to the inline-versus-mapped model above: a source table or column satisfies it either by being mapped to a business entity or attribute, or by carrying its own inline definition.

Marking a property Required decides what counts toward the completeness badges and filter covered below. On its own it stops nobody from doing anything. The page’s third part, Gates, is what turns a required-but-empty property into a real stop on a real action, such as generating a build: see Requiring and enforcing governance.

The governance panel

Wherever governance requirements are turned on, a Governance panel appears showing the properties your administrator enabled for that asset type. Each required property that is still empty shows a small Missing flag, and a header badge summarizes overall progress: not started, in progress, or complete.

The panel appears in three places:

Business entity sheet: open an entity from the modeling canvas or the glossary; the panel sits in the Governance section of the Details tab. It wraps the entity’s steward, owner, category, classification, and definition controls.
Business attribute governance popover: on an entity’s Attributes tab, each attribute has a small shield icon beside it. Select it to open a popover showing the panel wrapped around the attribute’s classification chips.
Source column edit dialog: open an existing, unmapped column for editing, not while adding a new one, and the panel appears inline with a definition box, classification buttons, and the Promote to attribute button described next.

Overriding and propagating classification on a mapped asset

A mapped column reads its classification from the attribute it governs, labeled From attribute {name} so you always know where the value came from. If one mapped column genuinely needs a different classification than the rest of its family, without touching the model, select Override classification next to the read-only badges. It opens the same chip editor the unmapped path uses; toggling a chip writes an override that takes precedence over the derived value immediately. Clear the override by untoggling every chip, and the panel falls back to showing the derived classification again. The definition is never overridable this way: it always stays owned by the entity or attribute.

To push a classification the other direction, from an entity or attribute out to every column mapped to it, its governance panel header offers a Propagate classification action next to Draft with AI whenever the entity or attribute already carries at least one classification. Selecting it opens the Ask AI panel and starts a run that proposes writing that classification onto each mapped source column, one change per column, reviewed the same way a drafted definition is: nothing changes until you apply it.

Promote a column to a business attribute

An unmapped source column’s inline draft is a starting point, not a destination. When you are ready to make it a real, governed part of your model, open the column for editing and select Promote to attribute in its governance panel.

Promoting does three things in one step:

Creates a new business entity and a business attribute under it, named from the column’s business name, or its column name if no business name is set.
Carries the column’s inline definition and classification over to the new attribute.
Maps the column to the new attribute and clears the column’s inline draft, since the attribute is now authoritative.

The new entity and attribute are skeletons, the same kind Map and match creates when no existing entity fits: fully governable right away, and wearing a Derived chip on the canvas until you open the entity sheet and select Adopt into model. Promoting from the column dialog always creates a fresh entity; if you would rather attach the column to an entity you already have, use Map and match instead.

Source tables do not have a promote shortcut. Map a table to an entity by hand on its Details tab, or through Map and match.

Completeness: badges and filters

Two existing surfaces pick up completeness once governance requirements are on:

Incomplete badge: the columns grid on a table’s Columns tab shows a small Incomplete badge next to any column missing a required property.
Governance filter: the entities directory offers a tri-state Governance filter, Either, Incomplete, or Complete, so you can pull up exactly the entities still missing required governance. Either is the default and applies no filtering.

Both read the same cached completeness that drives the panel’s header badge, so a change you make in the panel is reflected the next time either surface loads.

Drafting definitions with AI

Every governance panel that shows a definition field also offers a Draft with AI button next to the header badge. Selecting it opens the Ask AI panel and starts a run, using the asset’s name and any text already in the definition as context.

Like any AI skill run, you watch it work in the live progress view and it never writes anything on its own. When it proposes a definition, the panel shows a before-and-after comparison: your current text, if any, next to the proposed one. Leave it checked and select Apply to write the proposed definition, or Cancel to discard the proposal and leave your text exactly as it was. See Run an AI skill for how the live progress view and the Ask AI panel work in general.

Scanning for governance gaps with lint

Completeness badges and filters surface gaps one asset at a time. For a wider sweep across your whole business model, open the Ask AI panel from any Business page and select Run governance lint among the suggested skills. Like any AI skill run, you watch it work in the live progress view; when it finishes, a review page opens automatically with four sections: attributes carrying no definition, unmapped columns carrying no governance, names that break your organization’s naming convention, and entities marked deprecated that are still bound to live tables. Each row links straight to the place you would fix it. A lint run never changes anything on its own: it is a report to work from, the same read-only discipline as every diagnostic skill in DeltaVault.

Data governance overview: classifications, ownership, and the definition approval lifecycle a mapped entity or attribute carries
Map and match source metadata: mapping tables and columns, and the skeleton entities Promote to attribute also creates
Business entities: adopting a derived skeleton into your model
Classification settings: the schemes behind every classification button in the governance panel
Run an AI skill: how a skill run’s live progress and preview work