| name | mcp-engine-model-quality |
| description | Use when the user asks for a model audit, model quality review, scorecard, bad-practices or best-practices assessment, or a review of star-schema fit, relationships, DAX maintainability, VertiPaq/storage risk, metadata hygiene, governance signals, or validation gaps in a Power BI semantic model. For diagnosing one slow query, use mcp-engine-dax-performance; for Copilot or natural-language readiness, use mcp-engine-ai-readiness; to execute the remediation backlog, use mcp-engine-refactoring. |
PBI Model Quality
Use this skill to assess a connected Power BI semantic model and return a source-backed quality scorecard with prioritized recommendations. This is an assess-only workflow; do not apply model changes.
Start Here
- Confirm the current model context with SemanticOps MCP tools when needed.
- Gather metadata before querying data.
- Use
list_model, manage_dependencies, run_query, manage_tests, and manage_model_connection where available.
- Use
run_query only for small aggregated validation, performance analysis, VertiPaq/storage diagnostics, or access tests.
- Do not dump raw rows or sensitive values.
- Cite bundled Microsoft Learn and SQLBI source links for material findings.
Workflow
Assessment Areas
- Model shape and star-schema fit.
- Relationships and filter propagation risk.
- DAX and semantic layer maintainability.
- Storage and performance risk, including high-cardinality and unnecessary imported data.
- Metadata, naming, descriptions, display folders, and field exposure.
- Governance signals, including roles, sensitive-field exposure, and perspective-vs-security separation.
- Validation and test coverage.
Guardrails
- Do not call write operations from authoring or governance tools during the assessment.
- Treat unavailable Pro diagnostics, browse-only mode, policy denials, or missing tool capabilities as scope limitations, not model defects.
- Keep source-backed guidance nuanced; do not turn "generally recommended" practices into absolute rules when the source allows exceptions.
- Mark inferred findings with lower confidence unless tool evidence confirms them.
- End with concrete remediation steps and validation suggestions, not broad advice.
Output Standard
Return a compact quality assessment unless the user asks for raw detail:
- Executive score and quality band.
- Top 3 risks.
- Category scorecard.
- Findings grouped by critical, high, medium, and low severity.
- Prioritized remediation backlog.
- Validation/test recommendations.
- Source notes.
