Data governance

Control which workspaces reach which data, keep shared links accountable, and resolve duplicate data across sources. Grouped under Settings > Data Governance: Data Sources (bindings), Public Shares, and Data Overlaps.

Prerequisites

Admin role at the tenant level for binding, scoping, revoking shares, and resolving overlaps. The Classify and Extract Knowledge actions on a source need only the Editor role.

Data source bindings

A data source is connected once at the organization level, then bound to the workspaces that should query it. One connection, many audiences, you do not re-enter credentials per workspace.

Binding is more than on/off. Per workspace you can:

• Bind or unbind: The source, controlling whether that workspace sees it at all

• Scope with include and exclude patterns: A workspace sees only the schemas or tables it should

• Sync and diagnose: From the binding view, trigger a refresh, or check why a connection is unhealthy

This is how one warehouse serves many teams without each team seeing everything. A client workspace can be bound to only that client’s tables, even though the connection reaches the whole warehouse.

Note

Binding narrows what a workspace can query in Adverity Atlas. It works alongside your warehouse’s own row-level security, it does not replace it. The effective access is the intersection of the two.

A binding also carries a Query policy. Turning on Approved metrics only restricts agents to answering from active, approved measures on that source, refusing raw SQL and ad-hoc aggregates. See Measures.

Row filters and visibility

Open a source under Settings > Data Governance > Data sources and configure a workspace’s binding to control what that workspace sees inside the source, not just whether it sees it.

Control	What it does
Workspace labels	Key/value pairs the row filter can reference as workspace.labels.<key>. Shared by every binding on that workspace
Row filter	A predicate applied to every queryable asset under the source whose schema contains the referenced columns
Visibility	A read-only count of hidden assets and hidden fields. Change what is hidden in the schema browser

The row filter has two editors. Structured builds a list of clauses that are AND-composed: pick a column, an operator (equals, not equals, in list, not in list, matches, >, >=, <, <=, is null, is not null), then either a literal value or a workspace.labels.<key> reference. Raw JSON takes the predicate directly, and is the way to express OR, NOT, between, or nested groups. Leaving the JSON empty clears the filter.

Labels plus a label reference are what let one binding definition serve many workspaces. Filter on advertiser_name equals workspace.labels.brand, give each workspace its own brand label, and every workspace sees only its own rows through the same source.

Warning

A row filter only applies to assets whose schema actually contains the columns it references. A filter on a column a table does not have leaves that table unfiltered, so check the filter against every asset the workspace can reach.

Classify and extract knowledge

Opening a source also gives you two sections, one action each, in the order they should be performed:

1. Classification, described as “Map each column to a glossary concept”. Classify runs the semantic classification pipeline for the source, matching its columns to glossary concepts. Adverity Atlas asks you to confirm before it starts.

2. Knowledge, described as “Extract business knowledge, brands, markets, naming conventions”. Extract Knowledge reads the classified metadata and proposes knowledge items, brands, markets, and naming conventions, for you to confirm.

Run them in that order: extraction works from what classification produced.

Each button is disabled until the source is ready, and the reason is spelled out beneath it rather than only in a tooltip:

Action	Disabled until	Reason shown
Classify	The source has synced tables	Sync this source first, it has no tables to classify yet.
Classify	An AI provider is configured	Configure an LLM provider in Settings first.
Extract Knowledge	The source is bound to a workspace	Bind this source to a workspace first, extracted knowledge is stored per workspace.
Extract Knowledge	An AI provider is configured	Configure an LLM provider in Settings first.
Extract Knowledge	At least one column of the source has been profiled	Sync this source first, it has no profiled columns yet.

Both actions need the Editor role, not Admin. Both run in the background, and Adverity Atlas confirms the start with a message, “Semantic pipeline started” or “Business knowledge extraction started”. Extraction also needs a campaign-coded column to work from, and says so instead of starting if the source has none.

Every click on Classify is a run of its own, and the message tells you what that click did:

Message	What it means
Semantic pipeline started	Classification is running
A message that tables are being profiled first	Some tables need profiling before their columns can be classified. Classification starts once profiling finishes; follow both on the Tasks page
This data source has no tables yet, so there is nothing to classify. Sync it first, then click Classify.	Nothing was started. Sync the source, then try again
No LLM provider configured. Add one in Settings.	Nothing was started. Configure an AI provider first

A table that cannot be profiled does not stop the run. Classification continues over the tables that did profile.

A wide table can take more than one run to classify. Columns are saved as each batch finishes, and when a run nears its time limit it hands the remaining columns to a follow-up run, so you may see several classification runs for one table on the Tasks page. A run fails instead of continuing when it classified no new columns.

Public shares

When someone publishes an artifact to a public link, that link is listed here so it never becomes invisible.

1. Inventory, every active public share across the tenant, in one place.

2. Revoke, kill a link immediately. The URL stops working and the artifact returns to Restricted; its content is untouched.

Public share links expire on their own after a default lifetime. The sharer can revoke earlier, and an admin can revoke any link from this page, important when a link has been circulated more widely than intended.

Data overlaps

When a newly synced source duplicates data you already have, Adverity Atlas flags it here rather than silently querying both. Overlaps are detected three ways: matching asset names, field-name overlap of 50% or more, and overlapping sample values.

• Review: The two assets side by side and what they share

• Resolve: By choosing which source is authoritative, so answers stop drawing from both

• Dismiss: When the overlap is expected and no action is needed

Leaving overlaps unresolved risks double-counting, the same spend arriving from two sources and being summed. Resolving them keeps answers correct.

Permissions

Roles are cumulative, so the role named is the lowest one that can take the action.

Action	Required role
Query bound data sources	Analyst
Classify a source’s columns, extract knowledge from it	Editor
Bind or scope data sources	Admin
Revoke public shares	Admin
Resolve data overlaps	Admin

What’s next?

• Data sources, connect the sources you are binding here

• Workspaces, the workspaces you bind sources to

• Artifacts, where the public share links come from

Was this page helpful? Yes No