Skip to content

Fabric Data Agent Score Dimensions

To find what lowered a data agent's score and what to change, look up its dimensions on this page. In each dimension, you find the checks for one part of the data agent's setup, and each check tests one place where the setup might not follow Microsoft's published guidance.

Every score follows these rules:

  • A dimension that doesn't apply is left out of the score, not counted as passed. For example, a data agent grounded only on semantic models has no example queries to check.
  • A check that couldn't run is also left out.
  • If the data agent has no instructions, you see only that one finding for instructions. The other instruction checks run once there are instructions.
  • On the score card, you also see the Score limit from semantic models dimension. See Understand the score limit from semantic models.

To see the points each dimension cost, select Why this score? on the score card.

The Why this score? table of a data agent assessment, listing the points BI Pixie deducted from the score for each dimension. Data agent instructions lost 39 points. Published description, Runtime, Score limit from semantic models, Semantic model settings and Table selection lost no points.

Dimensions and checks

In the table, the two middle columns show what is checked without AI and what your AI is asked when you select Ask your AI.

Dimension Checked without AI Asked of your AI (when you select Ask your AI) Applies to
Data agent instructions
  • The data agent has instructions.
  • They are within Microsoft's limit of 15,000 characters.
  • Every data source the instructions name is one the data agent has.
  • A data agent with several data sources says which one answers which kind of question.
  • No sentence names a table, column or measure of one semantic model, because Microsoft Fabric ignores such guidance here, and no sentence spans several semantic models.
 Whether a sentence that names no table, column or measure is still about the data in one data source. Every data agent
Recommended format
  • The instructions use the five sections Microsoft recommends, Objective, Data sources, Key terminology, Response guidelines and Handling common topics, with content under each.
  • The instructions of each lakehouse, warehouse, KQL database, SQL database or mirrored database source use Microsoft's three sections, General knowledge, Table descriptions and When asked about.
This dimension has a low weight. A missing section suggests the format wasn't followed. It doesn't mean the data agent was told something wrong. 		Nothing Every data agent
Grounding matches the AI data schema The tables the data agent uses from a semantic model are the tables that semantic model's AI data schema defines. Nothing Semantic model sources
Table selection Every table the data agent selected still exists in the source. Nothing Every source whose tables can be read
Semantic model settings A semantic model source has no data source instructions and no data source description, because Microsoft Fabric doesn't read either for a semantic model. Nothing Semantic model sources
Published description The data agent has a published description. An AI that chooses which data agent answers a question reads this description. Nothing Every data agent
Runtime The data agent runs on Microsoft's preview runtime. Only the preview runtime has Microsoft's advanced query generation for semantic models and reads schema object descriptions. Nothing Every data agent
Data source setup
  • Each source has data source instructions and a data source description.
  • The instructions name tables the data agent selected and the source still holds.
  • The When asked about section states at least one rule.
  • The instructions say what to do rather than what to avoid.
  • The description says what the source contains and what it can answer.
  • The description tells this source apart from the data agent's other sources.
  • Whether the instructions define objects one by one, as a glossary.
  • Whether a line of the instructions defines one object, which belongs in that object's description instead.
 Lakehouse, warehouse, KQL database, SQL database and mirrored database sources
Example queries
  • Each source has example queries.
  • Each source has at least three, because Microsoft Fabric gives the data agent up to three of them for each question.
  • Every example has both a question and a query, and the questions are distinct.
  • Every number, date and quoted value in a question also appears in its query, which Microsoft calls the Mapping score.
  • Every table a query references is one the data agent selected and the source still holds.
  • At least one example has a comment, and at least one joins several tables.
  • On a KQL database, at least one example uses one of the database's functions.
  • The examples demonstrate the query patterns Microsoft asks for first.
  • Whether each question is clear, and whether it relates to its query, which Microsoft calls the Clarity and Relatedness scores.
  • Whether two examples answer the same question in different ways.
 Lakehouse, warehouse, KQL database, SQL database and mirrored database sources
Schema object descriptions
  • The data agent runs on the runtime that reads them.
  • Every selected table has a description.
  • Every column whose name reads as a code, a flag, a unit or a date has one.
  • A description says more than the name of the object it describes.
  • Similarly named objects are told apart.
  • The descriptions in the data agent still match the descriptions in the source.
  • When a Direct Lake semantic model reads this source and its reports have Pixies, the columns report viewers use most have a description.
  • Whether table descriptions state the entity, the level of detail and the purpose.
  • Whether column descriptions state the meaning, the unit or the expected values.
  • Whether a rule written in an object's description belongs in the data source instructions instead, because it applies across the source.
 Lakehouse, warehouse, SQL database and mirrored database sources
Data sources a semantic model already covers The data agent is grounded on both a semantic model and the lakehouse, warehouse or mirrored database that the semantic model reads through Direct Lake. Both hold the same data, but only the semantic model has the measures, relationships and descriptions. A question about that data can be sent to the copy without them. Nothing Data agents grounded on both
Tables report viewers never use The selected tables that no report viewer uses, next to the tables used most. Usage comes from the Pixies on reports over the Direct Lake semantic model built on the source. You find this under Fix on your own. It is never scored, because the data agent may still need such a table for a join or for a question no report answers. Nothing Lakehouse, warehouse, SQL database and mirrored database sources that a Direct Lake semantic model with tracked reports is built on

Microsoft's rules behind the checks

Most findings come down to five rules from Microsoft's documentation. When you open a dimension, you see the rule quoted with a Learn more link. The rules come from Data agent configurations, Semantic model best practices for data agent, Data agent configuration best practices and Add schema object descriptions.

  • What belongs in a data agent's instructions. "Data agent instructions should only include guidance that applies across all data sources configured in the agent, such as general response formatting preferences, cross-source routing rules, common abbreviations, tone, and so on."
  • What a data agent's instructions cannot do for a semantic model. "The DAX generation tool only refers to the AI instructions configured in Prep for AI of the semantic model. Data agent instructions aren't passed to the tool and are ignored when querying semantic models. For this reason, don't add semantic model specific instructions at the data agent level." Microsoft states the same limit for a semantic model source's own settings: "Unlike other data sources, data agent doesn't support data source instructions or descriptions for semantic models."
  • What a data source description is for. "A description should summarize what the data source contains, the types of questions it can answer, and any business-specific nuances that help distinguish it from other sources."
  • Where a definition belongs. "Place a definition in the schema object description when it applies to one object. Place it in data source instructions when it applies across the data source or affects multi-object query logic."
  • What editing an inherited description does. "By default, the data agent inherits available descriptions from the data source and updates them automatically. When you edit an inherited description in the data agent, the change applies only to the data agent. It doesn't overwrite the description in the source, and that object no longer receives description updates from the source."
Important: Two findings matter most, because in both cases the guidance you wrote has no effect:
  • Instructions or a description on a semantic model source. Microsoft Fabric doesn't support them for a semantic model.
  • Data agent instructions that explain how to query a semantic model. Microsoft Fabric never passes them to the tool that writes the DAX query.

Guidance about a semantic model's data belongs in that semantic model's AI Instructions, in Prep data for AI. Guidance that applies across sources, such as which source answers which question, the tone of answers and abbreviations, belongs in the data agent's instructions.

Ask your AI about wording

Some checks are about wording, such as whether a description says what its source contains or whether an example question is clear. To have your AI answer those checks and propose new instructions in the same round:

  1. Open the data agent's assessment, and go to the Data agent instructions card under Optimize data agent.
  2. Optionally, open How your AI works for you to see Which AI you are asking and which account or capacity it uses, How many questions the round asks at most, and What changes when it runs.
  3. Select Ask your AI.
  4. Review the proposed instructions on the same card. Nothing is written to the data agent until you apply them. See Optimize a data agent.

Asking your AI works as follows:

  • A data agent grounded only on semantic models has one such check, about its instructions. Each other source adds up to ten.
  • Your AI's verdicts are scored with the other checks, so the score can go up or down. A check your AI didn't answer is left out of the score.
  • No change figure is shown between an assessment that asked your AI and one that didn't, because the two answered different questions.
  • Answers are kept. In the next round, your AI is asked only about what changed, and in a round on an unchanged data agent, it is asked nothing.
  • A scheduled assessment never asks your AI, because asking your AI needs you signed in.
  • If the data agent has nothing for your AI to read, Ask your AI isn't offered.

Your AI receives the data agent's instructions, the instructions and descriptions of its data sources, its example queries, its schema object descriptions, and the names of the tables, columns and measures they mention. It never receives the data in your tables. For the full list, see AI setup.

Before your first round:

  • To set up your AI, select Set up your AI.
  • To let BI Pixie ask the BI Pixie data agent questions on your behalf, select Grant access. This permission is separate from the storage permission in Grant access to read a source.

Where the questions run and what they cost

  • Azure AI Foundry, OpenAI or Anthropic: the questions run against that account, and its charges apply. Nothing runs on your Fabric capacity.
  • The BI Pixie data agent: each question is answered by the data agent BI Pixie installed beside its dashboard, on the Fabric capacity of that workspace. See Microsoft's data agent consumption terms, which can change.

BI Pixie doesn't read your capacity's consumption, so it can't show what a round used. Before you ask, you can check the most questions a round can ask in the How many row. To measure your own data agent with the Fabric Capacity Metrics app, see How to Measure What a Fabric Data Agent Costs Per Question.

Related content