mcp-engine-model-quality

# 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 1. Confirm the current model context with SemanticOps MCP tools when needed. 2. Gather metadata before querying data. 3. Use `list_model`, `manage_dependencies`, `run_query`, `manage_tests`, and `manage_model_connection` where available. 4. Use `run_query` only for small aggregated validation, performance analysis, VertiPaq/storage diagnostics, or access tests. 5. Do not dump raw rows or sensitive values. 6. Cite bundled Microsoft Learn and SQLBI source links for material findings. ## Workflow - Read [model-quality-assessment-workflow](references/model-quality-assessment-workflow.md) for the inspection sequence, SemanticOps MCP tool usage, safety rules, and final output order. - Read [model-quality-scorecard](references/model-quality-scorecard.md) when scoring the model, assigning severity, formatting findings, and building the remediation backlog. - Read [model-quality-rulebook](references/model-quality-rulebook.md) for source-backed bad/questionable practice checks and recommended remediation language. ## 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: 1. Executive score and quality band. 2. Top 3 risks. 3. Category scorecard. 4. Findings grouped by critical, high, medium, and low severity. 5. Prioritized remediation backlog. 6. Validation/test recommendations. 7. Source notes.