semantic-view-optimization-patterns

name	semantic-view-optimization-patterns
description	Library of optimization patterns for dimensions, metrics, filters, relationships, and custom instructions. Use when root cause analysis identifies gaps needing targeted fixes.
sub_skills	["dimension_optimization.md","metric_optimization.md","filter_optimization.md","relationship_optimization.md","custom_instructions.md"]

Optimization Pattern Library

When to Load

• Root cause analysis (debug) identified semantic view gaps
• Audit identified structural issues to fix (e.g., missing relationships)

LLM Approach

These are patterns, not rigid rules:

• Understand the underlying principles
• Adapt to the specific semantic view and issue
• Use judgment for unusual cases
• Ask user when uncertain

Don't just copy examples - analyze the issue and apply the right fix creatively.

Load Strategy

Issue Type	Load Skill
Column not used correctly	dimension_optimization.md
Wrong aggregations	metric_optimization.md
Missing filter patterns	filter_optimization.md
Missing joins	relationship_optimization.md
Complex business logic	custom_instructions.md

Optimization Priority

1. Enhance Dimensions/Facts (descriptions, synonyms) → dimension_optimization.md
2. Add Metrics (aggregations) → metric_optimization.md
3. Add Filters (WHERE conditions) → filter_optimization.md
4. Add/Fix Relationships (joins) → relationship_optimization.md
5. Custom Instructions (last resort) → custom_instructions.md

Key Constraints

Cannot Add:

• Dimensions (physical columns - only enhance metadata)
• Facts (physical columns - only enhance metadata)

Can Add:

• Metrics (calculated aggregations)
• Filters (named WHERE clauses)
• Relationships (joins)
• Custom instructions

Prefer: Explicit model elements over custom instructions

References

• Example: example_semantic_view.yaml (valid field structures)
• Proto: semantic_model.proto (Dimension 105-134, Fact 174-199, Metric 307-325)

Guidelines

• DO NOT modify verified_queries
• ASK user for clarifications when ambiguous
• VALIDATE after each optimization
• ITERATE if first attempt doesn't work