Connection to semantic models

Power BI Audit supports two connection modes, accessible from the same dialog window upon startup.

Desktop mode (local)

When Power BI Desktop is open on your machine, Power BI Audit automatically detects active instances. No authentication is required; communication is entirely local. This mode is ideal for auditing a model currently under development before publication.

Service mode (XMLA)

This mode connects directly to a semantic model published on the Power BI Service via the XMLA protocol. It requires read/write XMLA endpoints (available on Fabric or Premium capacities) and authentication via your Microsoft account. This mode is suitable for auditing models in production.

Performance analysis (measuring actual execution times) is available in Service mode as long as your account has trace rights on the capacity.

Scoring system

The score is the centerpiece of Power BI Audit. It translates dozens of checks into a single, comparable, and actionable indicator.

Score by domain

Each domain starts with a score of 100. Findings from the same rule are grouped by severity level, and each group applies a capped penalty:

−15 pts max
per high rule
−8 pts max
per medium rule
−4 pts max
per low rule

The capping is important: even if a rule detects 100 problems of the same severity, the corresponding penalty will never exceed the maximum defined for that severity. This prevents a single highly repeated type of problem from entirely erasing a domain score. If the same rule generates findings of different severities depending on the case (for example, very slow measures in High and others moderately slow in Medium), each severity level is capped independently; the total penalty for this rule can then reach the sum of the relevant caps.

Weighted global score

The global score is the weighted average of the domain scores. The default weights are:

Domain Weight Justification
Modeling25 %The model structure is the foundation of everything; poor modeling impacts all other domains.
DAX25 %DAX measures are the heart of business logic; their quality determines maintainability and performance.
VertiPaq20 %The storage engine directly influences memory consumption and Fabric Capacity Units.
Power Query12 %M transformations impact refresh times and performance at the source.
Performance15 %Actual DAX execution times complement static analysis with factual measurements.
Governance3 %Documentation and conventions: important for maintainability, but less critical for performance.
If a domain cannot be analyzed (for example, Performance if trace rights are not available), its weight is automatically redistributed to the other analyzed domains. The total always remains 100%.

Analysis domains

Power BI Audit covers six distinct domains. Each domain groups a coherent set of rules checked automatically on your model. See the audit rules page for the full detail of each rule.

Modeling

Checks the structural quality of your semantic model: star schema, identification of fact and dimension tables, relationship management (bidirectional, inactive, many-to-many), date table, calculated columns in facts, etc. This is the most fundamental audit; a poorly structured model degrades all other indicators.

DAX

Static analysis of all DAX measures in the model. Power BI Audit detects anti-performance patterns (FILTER(ALL()), nested iterators, IFERROR...), bad practices (missing DIVIDE, unused variables, implicit measures), and risks of overly deep dependencies between measures.

VertiPaq

Analysis of the column-by-column storage engine: high cardinality, insufficient compression, unused columns not hidden, overall model size, unsuitable partitioning, missing incremental refresh on large tables. These metrics are directly correlated with Fabric Capacity Unit consumption.

Performance

Actual execution of each DAX measure on the model with a cold cache before each measure. Power BI Audit measures total duration, Formula Engine (FE) share, Storage Engine (SE) share, CPU time, and the number of SE queries. This domain identifies objectively slow measures beyond what static analysis alone can detect.

Governance

Checks model documentation (descriptions of tables, columns, and measures), naming conventions, the presence and quality of row-level security (RLS) roles, and best practices around KPI and measure formats.

Power Query

Analyzes M scripts for each query to detect transformations that should ideally be performed directly at the SQL source: joins, sorts, groupings, column selections, calculated columns. These transformations performed on the Power Query side consume more resources and slow down refreshes.

DAX performance analysis

Performance analysis is the most advanced feature of Power BI Audit. It goes beyond static analysis by actually measuring the behavior of your measures on the engine.

How it works

For each DAX measure in the model, Power BI Audit:

  1. Clears the engine cache (cold cache) for measures unbiased by caching.
  2. Executes the measure via a dedicated DAX query.
  3. Collects XMLA trace metrics: total duration, FE duration, net SE duration (including parallelism calculation), CPU time, number of SE queries, and SE cache hits.

Metrics collected

MetricDescription
Total durationComplete end-to-end execution time, in milliseconds.
FE durationTime spent in the formula engine (FE). A high FE indicates poorly written iterators or complex dependencies.
SE durationTime spent in the storage engine (SE), corrected for parallelism. A high SE indicates scans of high-cardinality columns.
CPU timeTotal query processor consumption.
SE queriesNumber of queries sent to the storage engine. A high number may indicate iterators generating too many sub-queries.
SE cache hitsNumber of SE queries served from the cache. In a cold cache, this number is normally 0 for the first execution.

Thresholds and severity

The default slowness threshold is 1,000 ms (configurable). A measure exceeding this threshold generates a finding:

  • High if the duration exceeds 3 times the threshold (> 3,000 ms by default).
  • Medium if the duration exceeds the threshold without reaching this multiplier.
Performance analysis increases the total audit duration, proportional to the number of measures in your model. For a model with 200 measures and a timeout of 5 seconds per measure, expect up to 15 to 20 minutes in the worst case. In practice, the vast majority of measures respond in less than one second.

Light and dark themes

Power BI Audit offers two visual themes: a dark theme (enabled by default) and a light theme. Switching is performed via the toggle button in the main toolbar. Your preference is saved between sessions.

Results export

Audit results can be exported as an Excel file (.xlsx) from the main toolbar. The export contains two sheets:

  • Results, a summary of scores by domain and global score.
  • Findings, a complete list of all findings with their domain, severity, rule ID, concerned object, and recommendation.

See the results export page for full details.