Measures

A concept says what a term means. A measure says how to compute it on one specific table. “Spend” is the concept; SUM(cost_micros) / 1000000 on your Google Ads spend table is the measure. One concept can have many measures, one per source asset, so the same business term resolves correctly whichever table an answer is drawn from.

Important

Available on demand. Measures may not be enabled in your workspace by default. Contact your Adverity account manager to activate them.

Prerequisites

• A glossary concept to attach the measure to

• A connected data source bound to the workspace, with its assets discovered

• Editor role to author, activate, approve, or deprecate a measure

Key concepts

Term	What it means
Status	Draft, Active, or Deprecated. Only Active measures can be queried
Approval	A separate flag from status. A measure can be approved without being active, and active without being approved
Origin	Where the measure came from: User, Agent, Discovered, Derived, or Seeded
Probe	The result of a validation query run against your warehouse: Valid, Failed, or Unvalidated
Scope	A measure is tenant-wide by default, or scoped to a single workspace as an override

Note

A measure that fails its warehouse probe can never leave Draft. This is deliberate: an implementation that does not run is not one Adverity Atlas will answer from.

Where measures appear

Measures live inside the Glossary, they have no separate top-level area:

• On a concept: Open a concept from the Glossary and look for the Implementations panel, which lists every measure bound to that concept

• In the review queue: From the Glossary, click Measures to open the full list, described in-product as “Review and approve the executable metric implementations proposed for your glossary concepts”

In Chat, Adverity Atlas lists and queries only the measures the workspace can use: tenant-wide measures and the workspace’s own overrides, on a data source actively bound to the workspace, on a table the binding does not hide, and whose concept the workspace’s glossary shows (see Concept layers).

Reviewing proposed measures

Measures are proposed automatically after classification, so the usual job is reviewing them rather than writing them.

1. Open the queue: Glossary, then Measures

2. Filter to what needs a decision: Scope and concept or asset

3. Approve and activate: One at a time, or in bulk

In detail:

1. Open the workspace Glossary and click Measures. The page opens on the Draft scope, which is the review queue.

2. Switch scope with the Draft / Active / All toggle. Narrow further with the All concepts and All assets selects. The count to the right reads “N measure(s)”.

3. Read the row. Source is the asset the measure runs on, SQL is the aggregate, Origin is where it came from, and Probe shows whether it has validated against the warehouse.

4. Click Approve & activate on a draft. This is the primary action because approval on its own does not make a measure queryable, it has to be active too.

5. To handle several at once, tick the checkboxes and click Approve & activate selected. Adverity Atlas reports how many were approved and activated, and calls out any that were approved but could not be activated because the probe failed.

The remaining actions are available where they apply: Activate on a measure that is not yet active, Approve on one that is not yet approved, and Deprecate on an active one.

Tip

If the queue is empty, the empty state says so. Draft measures proposed after classification appear under the Draft scope, so check the scope toggle before concluding there is nothing to review.

Authoring a measure

When no proposal covers what you need, write the implementation yourself.

1. Open the concept from the Glossary.

2. In the Implementations panel, click Add implementation. A panel opens titled “Add implementation”.

3. Fill in the fields below.

4. Click Create measure. The measure is created as a Draft and appears in the Implementations panel.

Field	Description
Source asset	The table the measure runs on. The picker reads “Select an asset…”	Required
Measure name	Defaults to the concept name if you leave it blank	Optional
Aggregate SQL	A Spark aggregate over the asset’s columns, for example SUM(cost_micros) / 1000000	Required
Row filter (optional)	A row predicate compiled into the aggregate, for example status != 'refunded'	Optional
Description	Free text explaining what this implementation covers	Optional

To change an existing measure, click Edit on its row, adjust the name, SQL, filter, or description, and click Save changes. The source asset and the concept are fixed once created, so a measure that belongs on a different table is a new measure.

Warning

Editing the SQL or the row filter re-runs the warehouse probe. If the new expression does not validate, the measure records the failure and cannot be activated until it does.

Validation against your warehouse

Adverity Atlas validates a measure in two ways before trusting it. The expression is checked against the columns that actually exist on the asset, and a probe query is run against the warehouse that returns no rows. The probe is what fills in the Probe column: Valid once it has succeeded cleanly, Failed with the warehouse’s own error when it has not, and Unvalidated before it has been tried.

High-confidence measures that Adverity Atlas proposes for you are probed and activated automatically once they come back clean, so the review queue holds the ones that genuinely need a decision rather than every proposal.

Restricting a source to approved metrics

Measures make a stricter governance mode possible. On a data source binding, turn on Approved metrics only under Query policy and agents can answer from active, approved measures on that source alone. Raw SQL and ad-hoc aggregates are refused.

This is the setting to reach for when a source carries data that must only ever be reported one way. Configure it per workspace on the binding, see Data governance.

Permissions

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

Action	Required role
View measures and the review queue	Viewer
Add an implementation	Editor
Edit, activate, approve, or deprecate	Editor
Set a source binding to approved metrics only	Admin

What’s next?

• Adding concepts, the concept a measure implements

• Data governance, bind a source and set its query policy

• Chat, ask a question and see the measures answer it

• Data sources, the assets measures are written against

Was this page helpful? Yes No