Reference
Audit rules
Complete list of rules checked by Power BI Audit, classified by domain, with their functional description and severity.
High Max penalty −15 pts
Medium Max penalty −8 pts
Low Max penalty −4 pts
Modeling
Semantic model structure - schema, relationships, types, tables
25 %
of global score
| Rule | Description | Severity | Recommendation |
|---|---|---|---|
| MOD001 | Snowflake schema - dimension linked to another dimension (snowflake) | High | Flatten nested dimensions to obtain a pure star schema. Snowflake schemas increase DAX query complexity and degrade VertiPaq performance. |
| MOD002 | Fact tables not identifiable in the model | Medium | Ensure your fact tables are recognized as such: they must be on the N side of a relationship with at least one dimension table. |
| MOD003 | No dedicated date table present | High | Create a dedicated date table and mark it as such. DAX time intelligence only works correctly with a well-configured date table. |
| MOD004 | Date table present but not marked IsDateTable | Medium | Mark your date table via the "Mark as date table" option in Power BI Desktop. This activates VertiPaq optimization and DAX time intelligence validations. |
| MOD005 | Uncontrolled many-to-many relationship detected | High | M:N relationships are rarely desirable. Introduce a junction table or review the modeling to switch to classic 1:N relationships. |
| MOD006 | Bidirectional relationships without apparent justification | Medium | Bidirectional filters can cause unexpected behavior and degrade performance. Use CROSSFILTER() in DAX if bidirectional filtering is only needed in certain measures. |
| MOD007 | Inactive relationships present without use of USERELATIONSHIP() | Low | Unused inactive relationships clutter the model. Delete them if they are not activated by any measure via USERELATIONSHIP(). |
| MOD008 | Isolated tables - not linked to the rest of the model | Medium | Each table must be linked to the model via at least one active relationship. Isolated tables cannot filter other tables and are often a sign of modeling errors. |
| MOD009 | Foreign keys not hidden in fact tables | Low | Foreign key columns (ID) in fact tables should be hidden in the report view. They are only useful for relationships and have no direct analytical value. |
| MOD010 | Date columns with incorrect data type (not Date or DateTime) | Medium | Date columns must have the Date or DateTime type. A Text or Integer type prevents time intelligence functions and degrades filtering performance. |
| MOD011 | Calculated columns present in a fact table | High | Calculated columns in fact tables are evaluated at each refresh and stored in memory column by column. Move these calculations to Power Query or the SQL source. |
| MOD012 | No dedicated measure table | Low | Group all your measures into one or more dedicated tables (without data). This significantly improves model readability and navigation for end users. |
| MOD013 | Calculated tables that could be replaced by M transformations | Medium | DAX calculated tables are fully recalculated at each refresh. Prefer Power Query (M) transformations for static derived tables - they do not consume DAX engine memory. |
| MOD015 | Unused columns not hidden in the report view | Low | Columns not used in visuals, measures, or relationships should be hidden (or deleted) to lighten the model and simplify the end-user experience. |
| MOD016 | Primary key not identifiable in dimension tables | Low | Each dimension must have a column identified as the primary key (1 side of the relationship) to ensure referential integrity and allow Power BI to optimize joins. |
| MOD017 | Excessive number of columns in a fact table (configurable threshold) | Medium | Wide fact tables (many columns) degrade VertiPaq performance. Review the necessity of each column - descriptive attributes belong in dimensions, not facts. |
| MOD018 | Primary key column (side 1) not hidden in a dimension | Low | The side 1 key column of a relationship can be hidden if it has no direct use in reports. It remains functional for relationships even when hidden. |
| MOD019 | Suspect table - present on both side 1 and side N of relationships | Medium | A table that is both on the N and 1 side of different relationships is often a sign of hybrid modeling to be reviewed. Check if this is an intentional choice and document it. |
| MOD020 | Auto date/time enabled - LocalDateTable detected | High | The auto date/time feature generates hidden date tables for each date column, significantly increasing model size. Disable it and use your own date table. |
| MOD021 | Fact → Fact relationship detected | High | Direct relationships between fact tables are a classic modeling error. Introduce a shared dimension table (degenerate dimension or bridge dimension) to link your fact tables. |
| MOD022 | One-to-one (1:1) relationship detected | Medium | 1:1 relationships often indicate two tables that should be merged. Check if this separation is intentional or a migration artifact. |
| MOD023 | Snowflake schema - dimension linked to other dimensions (cascade) | High | Similar to MOD001 but for deep dimension cascades. Flatten the hierarchy to return to a star schema. Use Power Query to denormalize if necessary. |
| MOD024 | Do not use floating point data types | Medium | Replace the Double type with Decimal for monetary or financial values, or with Int64 for integers. The Double type should only be used for scientific calculations where absolute precision is not required. |
| MOD025 | Set data coverage definition on DirectQuery partition of hybrid table | Medium | Define a DataCoverageDefinition expression on the DirectQuery partition to tell the engine the data range covered by the Import partition. |
| MOD026 | Dual mode irrelevant without DirectQuery | Medium | Convert Dual partitions to Import if the model is purely in Import mode, or add a DirectQuery source if Dual mode is intentional. |
| MOD027 | Set dimension tables to Dual mode when using DirectQuery on fact tables | Medium | Switch Import dimension tables to Dual mode when the model contains hybrid tables. This allows the engine to use in-memory data or send the query to the source depending on the context. |
| MOD028 | Do not summarize numeric columns | Medium | Set SummarizeBy = 'None' on all visible numeric columns and create explicit DAX measures for all required aggregations. |
| MOD029 | Add data category for geographic columns | Low | Set the DataCategory property on the column to the appropriate geographic value (Country, City, Continent, Latitude, Longitude) in Power BI Desktop. |
| MOD030 | Percentages should be formatted correctly | Medium | Use the standard format '#,0.0%;-#,0.0%;#,0.0%' for all percentage measures to ensure consistent display of positive, negative, and zero values. |
| MOD031 | Calculation items must have an expression | High | Define a valid DAX expression on all Calculation Items. If the Calculation Item is no longer needed, remove it from the Calculation Group. |
| MOD032 | Data columns must have a source column | High | Check the column definition in Power Query and ensure the SourceColumn property points to a valid column in the data source. |
DAX
Static analysis of DAX measures - patterns, practices, dependencies
25 %
of global score
| Rule | Description | Severity | Recommendation |
|---|---|---|---|
| DAX001 | Use of FILTER(ALL(...)) | High | Replace FILTER(ALL(Table), condition) with CALCULATETABLE(ALL(Table), condition) or restructure the filter in CALCULATE() directly. |
| DAX002 | Nested iterators (SUMX, AVERAGEX… in another iterator) | High | Refactor with variables or intermediate tables to reduce the cardinality of evaluation contexts. |
| DAX003 | Division without DIVIDE() | Medium | Replace A / B with DIVIDE(A, B) to automatically handle divisions by zero without raising an error. |
| DAX004 | Use of COUNTROWS(FILTER(...)) | High | Replace with CALCULATE(COUNTROWS(Table), condition). The version with FILTER iterates all rows; the CALCULATE version relies on the VertiPaq index. |
| DAX005 | IF() without explicit ELSE branch | Low | An IF() without ELSE implicitly returns BLANK(). Make the return explicit to avoid unexpected behavior in totals. |
| DAX006 | Variables declared but not used in the measure | Low | Delete unused variables. |
| DAX007 | Empty measure - empty body or containing only a comment | Medium | Delete empty measures or complete their implementation. |
| DAX008 | Use of IFERROR() | Medium | Prefer DIVIDE() for divisions and handle errors at the source in Power Query. |
| DAX009 | Nested CALCULATE() | Medium | Refactor by combining filter modifiers into a single CALCULATE(). |
| DAX010 | Time intelligence functions used without marked date table | High | Functions like TOTALYTD() and SAMEPERIODLASTYEAR() require a table marked as IsDateTable. |
| DAX011 | Reference to a column without aggregation in a measure | High | Always add an aggregation function (SUM, MAX, etc.) to avoid errors outside of row context. |
| DAX012 | Excessive dependency depth between measures | Medium | Flatten the hierarchy or group intermediate calculations to make debugging easier. |
| DAX013 | Use of RELATED() in cascade across multiple tables | Medium | Consider denormalizing the column directly in Power Query. |
| DAX014 | Implicit measures present with calculation groups | High | Convert all aggregations into explicit measures when using Calculation Groups. |
| DAX015 | Calculation group items without description | Low | Add a description to each Calculation Item for end users. |
| DAX016 | CONCATENATEX() on a large table | High | Limit its use to small reference tables to avoid extreme slowness. |
| DAX017 | SUMMARIZE() with extension columns | Medium | Use ADDCOLUMNS(SUMMARIZE(...)) instead. |
| DAX018 | ADDCOLUMNS() without CALCULATE() on expressions | Medium | Use CALCULATE() to transition row context into filter context. |
| DAX019 | Use TREATAS instead of INTERSECT | Medium | Replace INTERSECT(...) with TREATAS(table, column1, column2, ...) when the goal is to propagate a filter across tables. |
| DAX020 | EVALUATEANDLOG not in production | Medium | Remove EVALUATEANDLOG() from all measures before going to production. This function is reserved for local debugging. |
| DAX021 | Measure is a direct reference to another measure | Medium | Remove the redundant measure and use [OtherMeasure] directly where it is referenced. |
| DAX022 | No two measures should have the same definition | Medium | Remove the duplicate measure and keep a single version. |
| DAX023 | Avoid constant addition to division result | Medium | Review the calculation logic. If an offset is intentional, encapsulate it in a VAR variable to make the intent explicit. |
| DAX024 | Avoid 1-(x/y) syntax, use 1 - DIVIDE(x, y) | Medium | Replace the '1 - x/y' pattern with '1 - DIVIDE(x, y)' to benefit from division-by-zero protection. |
| DAX025 | Filter column values with proper syntax | Medium | Replace CALCULATE([Measure], FILTER('Table', 'Table'[Column] = value)) with CALCULATE([Measure], 'Table'[Column] = value) when the filter targets a single column. |
| DAX026 | Column references should be fully qualified | High | Always qualify column references with their table name: 'Table'[Column] or Table[Column]. |
| DAX027 | Measure references should be unqualified | High | Replace 'Table'[MyMeasure] with [MyMeasure] in all DAX expressions. |
| DAX028 | Avoid SUM or AVERAGE on a string column | High | Check the data type of the referenced column. Convert to a numeric type or use COUNTROWS / DISTINCTCOUNT as appropriate. |
| DAX029 | Inactive relationships never activated | Medium | Check whether the inactive relationship is intentional. If it is never used, remove it. |
| DAX030 | Calculation groups with no calculation items | Medium | Either complete the Calculation Group by adding items, or delete the table. |
Governance
Documentation, naming conventions, RLS security
3 %
of global score
| Rule | Description | Severity | Recommendation |
|---|---|---|---|
| GOV001 | Model description missing | Low | Add a global description to the semantic model explaining its purpose and scope. |
| GOV002 | Tables without description | Low | Document each table with a description indicating its content and granularity. |
| GOV003 | Columns without description | Low | Document key columns, especially those exposed to end users. |
| GOV004 | Generic names detected (e.g., Column1, Table2) | Medium | Rename objects with explicit business names. |
| GOV005 | Measure naming convention not respected | Low | Adopt a uniform convention (e.g., [Total Sales]) and apply it to all measures. |
| GOV006 | Table naming convention not respected | Low | Standardize table names with a clear prefix (e.g., dim_, fct_). |
| GOV007 | No RLS role present on a model with sensitive data | High | Implement row-level security (RLS) as soon as the model contains restricted data. |
| GOV008 | RLS role with empty filter | High | An RLS role with an empty filter or TRUE() grants access to all data. Correct this immediately. |
| GOV009 | Unbalanced parentheses in RLS filter expression | High | Check and correct the DAX syntax of the role expression. |
| GOV010 | RLS roles without description | Low | Document each RLS role: targeted users and restricted data scopes. |
| GOV011 | KPI measures without status indicator or target defined | Low | Complete KPI definitions with a target value and status thresholds. |
| GOV012 | Measures exposed to users without description | Medium | Visible measures must have a description explaining their calculation and unit. |
| GOV013 | Measures without defined FormatString | Low | Define a display format (currency, percentage, etc.) for each measure. |
| GOV014 | Generic column name | Medium | Rename the column with an explicit business name. |
| GOV015 | Generic measure name | Medium | Rename the measure with an explicit business name. |
| GOV016 | Check if dynamic RLS is necessary | Low | Verify that dynamic RLS is actually needed. Consider static RLS for small, stable user groups. |
| GOV017 | Limit RLS logic complexity | Medium | Simplify the RLS filter by pre-calculating values in a dedicated column. |
| GOV018 | Objects should not start or end with a space | High | Rename the object by removing leading and trailing spaces. |
| GOV019 | First letter of objects must be capitalized | Low | Rename the object by capitalizing the first letter for consistent readability. |
| GOV020 | Object names must not contain special characters | Medium | Remove invisible control characters (tabs, carriage returns, newlines) from names. |
| GOV021 | Mark primary key columns | Low | Enable the IsKey property on all columns on the 'One' side of relationships in dimension tables. |
VertiPaq
Columnar storage engine - memory, compression, partitions
20 %
of global score
| Rule | Description | Severity | Recommendation |
|---|---|---|---|
| VPQ001 | High cardinality columns detected | High | Hide or delete high cardinality columns if they are not used in visuals or measures. |
| VPQ002 | Insufficient compression ratio on some columns | Medium | Evaluate if the column is necessary and optimize its data type. |
| VPQ003 | Hash encoding where Value encoding would be more efficient | Medium | Check that your numeric columns have the correct data type (Integer vs Decimal). |
| VPQ004 | Unused columns not deleted from the model | High | Delete columns not used in any visual, measure, or relationship. |
| VPQ005 | Excessive total model size | High | Analyze the heaviest tables and optimize column by column. |
| VPQ006 | Fact tables with too many columns | Medium | Reduce the number of columns by moving attributes into dimension tables. |
| VPQ007 | Too low ratio of useful data vs total volume | Medium | Eliminate unnecessary columns or tables. |
| VPQ008 | Large tables not partitioned | High | Implement partitioning by date to allow incremental refreshes. |
| VPQ009 | Disproportionate partition sizes | Medium | Rebalance partitions to optimize parallelism during refreshes. |
| VPQ010 | Empty partitions detected | Low | Delete or merge empty partitions. |
| VPQ011 | Incremental refresh missing on large tables | High | Configure incremental refresh for all tables exceeding the critical size threshold. |
| VPQ012 | High referential integrity violations | High | Clean data at the source to guarantee referential integrity. |
| VPQ013 | Excessive number of segments per column | Medium | Check the data type and cardinality of the column. |
| VPQ014 | Tables with columns not resident in memory (DirectLake) | Medium | Force residency for frequently queried columns to avoid falling back to DirectQuery. |
Performance
Actual execution of DAX measures with XMLA trace analysis
15 %
of global score
| Rule | Description | Severity | Recommendation |
|---|---|---|---|
| PER001 | Slow DAX measure - total FE + SE duration exceeds configured threshold (default: 1,000 ms) | High / Medium | Simplify the measure, add variables, and reduce the cardinality of involved columns. |
Power Query
Analysis of M queries - transformations foldable to SQL source
12 %
of global score
Rules PQY001 to PQY008 apply only to queries where the source is a SQL database. PQY009 applies to non-SQL sources.
| Rule | Description | Severity | Recommendation |
|---|---|---|---|
| PQY001 | Join performed in Power Query (Merge) on a SQL source | High | Perform joins directly in SQL at the source (JOIN clause). |
| PQY002 | Reference to another Power Query query (Reference) | High | Verify if query folding is maintained; otherwise, refactor into a single SQL query. |
| PQY003 | Sort performed in Power Query (Table.Sort) on a SQL source | High | Move the sorting to the SQL query (ORDER BY). |
| PQY004 | Column selection in Power Query (Table.SelectColumns) on a SQL source | High | Select only necessary columns directly in the SQL query. |
| PQY005 | Grouping performed in Power Query (Table.Group) on a SQL source | High | Perform the grouping in SQL (GROUP BY). |
| PQY006 | Adding calculated column in Power Query (Table.AddColumn) on a SQL source | High | Add the calculated column directly in the SQL query or a view. |
| PQY007 | Nested join in Power Query (Table.NestedJoin) on a SQL source | High | Replace with LEFT JOINs in your source SQL query. |
| PQY008 | Use of Navigate to access a table in a SQL source | High | Define the SQL query directly in the Source step (Value.NativeQuery). |
| PQY009 | Non-SQL data source detected for high data volume | Medium | Migrate to a relational database (SQL Server, PostgreSQL, etc.) for better performance. |
| PQY010 | The number of steps in the M query exceeds the configured threshold (default: 10). | High | A high number of M steps increases the risk of slow refreshes and excessive Capacity Unit consumption on large volumes. Simplify the query or push transformations back to the source. |
| PQY011 | Use of Table.Unpivot or Table.Pivot detected in the Power Query query. | Medium | These functions consume significant resources during refreshes, especially on large data volumes, which can considerably increase the Capacity Unit consumption of Data Refresh operations on the Fabric capacity. Consider restructuring the table at the source or limiting the volume processed. |
| PQY012 | Use of SharePoint.Files detected in the Power Query query. | High | SharePoint.Files performs a recursive search across the entire SharePoint site, which is costly in terms of performance. Use SharePoint.Contents to extract files located in a specific folder or location, and reserve SharePoint.Files for cases where a genuine site-wide search is required. |