Semantic layer has two tabs for work waiting on a decision: Suggestions, where proposed metrics and questions wait to be added to your vocabulary, and Questions, where a saved question waits to be verified as one your tenant stands behind. Anyone who can read the semantic layer can see both tabs; only someone with the certify privilege can act on a row. A suggestion can come from a teammate who noticed a metric was missing, or from an agent that filed one while answering a question. Reading and filing needs only read access — deciding is what’s gated.

Reviewing a suggestion

1

Open Suggestions

Go to Semantic layer → Suggestions. The list opens filtered to Pending; use the status dropdown to see Accepted, Accepted with edits, Dismissed, or All statuses.
2

Read the proposal

Each row shows its Kind (Metric or Verified query), what it Proposes, who proposed it, and when it was filed. A metric proposal names the model it would land on; a question shows the question and the query behind it.
3

Accept or Dismiss

Click Accept to add the metric to its model’s draft, or to create a proposed verified query. Click Dismiss to decline it. Both ask for a required note before they take effect.
What each decision actually does: After you confirm, a message tells you what happened: “Accepted — it is on the model’s draft head, awaiting a proposal,” “Accepted — it is now a proposed verified query,” or “Dismissed — the author can see your note.” If you arrived through an AI link, the focused view explains that review happens in the console and nothing is sent to the chat. Both Accept and Dismiss finish the focused task after your decision succeeds. The banner names the decision and prompts you to return to your AI and tell it you have finished.
If the model a suggestion targets is already awaiting review when you accept the suggestion, accepting it withdraws that review — the definition the reviewer was looking at just changed.
You can only accept a suggestion as it stands or dismiss it. Editing a suggestion’s contents before accepting it isn’t available yet.
If a suggestion is on a model bound to a table you can’t see, it doesn’t appear in your list at all. When nothing has ever been proposed, the list says Nothing has been proposed yet. When a status filter matches nothing, it says Nothing with that status — clear the filter to see the rest of the queue.

Verifying a question

A verified query is a question your tenant has agreed how to answer: the question in someone’s own words, plus the exact query that answers it. A question lands here after someone saves it from a metric’s Ask this panel, or after a suggestion for it is accepted. Save as verified query is offered only once the model has a certified head. A verified query binds to your certified vocabulary, so a model that has never been certified has no metric for it to resolve against — until then the button stays disabled with a Certify v1 first hint. Propose and certify the model, then save the question.
1

Open Questions

Go to Semantic layer → Questions. Every row here is waiting to be verified.
2

Read the question

Each row shows the Question, what it Asks for, who saved it, and when.
3

Verify or Deprecate

Click Verify to check the question against today’s vocabulary and mark it as one of your tenant’s agreed answers. Click Deprecate to retire it without verifying. Both ask for a required note.
After you confirm, a message tells you what happened: “Verified — it now appears on its models’ pages as one of the agreed answers,” or “Retired — it stays readable and stops being offered as an example.” Once verified, a question appears on the page of every model it draws on, as a one-click question on a browsed table in Explore, and as an example an agent sees when it reads that model over MCP. When nothing is waiting, the list says Nothing is waiting to be vouched for.

Semantic layer overview

What models, metrics and concepts mean, and how they move from draft to certified.

Create models and metrics

Where a certified metric comes from before it ever reaches this queue.