Standard Skill: Assess the Semantic Layer
Assess the Semantic Layer is a Standard Skill: it judges how well the Semantic Layer serves one Data Source and saves that judgement as its Quality score. The agent uses it only in the Conversation that Assess in the Data Studio starts. It runs there on its own: it asks you nothing, proposes nothing and uses only tools that need no Approval.
What the agent does
Section titled “What the agent does”- It reads the Data Model and the entries linked to the Data Source, and the tables of the Data Source.
- It reads the Query Log of the last 90 days: the questions asked often, the queries that failed or were cut, and the answers rated down. It never names who asked.
- It judges three things, none of them by counting: how much of what users query the Data Model covers, with descriptions, relationships and measures; whether the Terms, Rules and Verified Queries answer what users ask, and whether entries contradict each other; and whether an entry would have led the agent to the right query for the frequent questions, failures and answers rated down. More entries alone don’t raise the score.
- It picks the score between 0 and 100 against fixed anchors, from no Data Model and next to no entries (0 to 20) to users’ questions answered from verified entries (81 to 100).
- It saves the score with its reasons in three parts: Covered, Missing (most valuable first, each with the kind of entry that would fix it) and Failing questions. Then it answers in two or three sentences.
The agent saves a score even when it could read little, and says so in the reasons. If the Conversation ends without a saved score, the assessment counts as failed and the previous score stays.
Improve the Semantic Layer starts from these reasons.