semantic-data-bundle

# Semantic Data Bundle Agent uses this bundle for semantic data work spanning terminology, ontology structure, semantic models, graph patterns, and natural-language query translation. ## Activation | Prompt Signal | Use Bundle | Primary Route | |---|---|---| | "ontology", "taxonomy", "class model", "controlled vocabulary" | Yes | `ontology-design-patterns` | | "semantic model", "Fabric model", "Power BI model", "metrics layer" | Yes | `semantic-layer-fabric` | | "business glossary", "term catalog", "definition governance" | Yes | `business-glossary-metadata` | | "term harmonization", "synonym cleanup", "canonical naming" | Yes | `terminology-normalization` | | "entity relationships across domains", "graph traversal", "knowledge graph" | Yes | `knowledge-graph-patterns` | | "natural language to SQL", "NLQ", "query translation" | Yes | `nlp-to-query-translation` | | "modeling standards", "naming standards", "semantic consistency" | Yes | `semantic-modeling-standards` | ## When to Use | Request Pattern | Use | |---|---| | Multi-step semantic architecture or governance request | Use this bundle | | Decision needed across glossary, ontology, semantic model, and graph layers | Use this bundle | | Fabric + Purview + graph integration request | Use this bundle | ## When Not to Use | Request Pattern | Route | |---|---| | Single glossary curation task only | `business-glossary-metadata` | | Single Fabric semantic model task only | `semantic-layer-fabric` | | Single NLQ prompt translation task only | `nlp-to-query-translation` | ## Required Inputs | Input | Required | Notes | |---|---|---| | Data domain | Yes | Finance, operations, HR, supply chain, or other bounded scope | | Consumer pattern | Yes | Analytics, governance, search, application, or conversational query | | Platform context | No | Fabric, Purview, graph database, SQL engine, BI client | | Output target | No | Glossary, ontology, semantic model, graph schema, NLQ interface | ## Coverage Matrix | Concern | Specialist Skill | Primary Output | |---|---|---| | Class, property, and ontology reuse patterns | `ontology-design-patterns` | Ontology structure | | Fabric-ready semantic layer design | `semantic-layer-fabric` | Measures, dimensions, model contracts | | Business term stewardship and metadata | `business-glossary-metadata` | Governed glossary | | Canonical term alignment and synonym reduction | `terminology-normalization` | Normalized vocabulary | | Multi-hop relationship modeling and graph traversal design | `knowledge-graph-patterns` | Graph schema and patterns | | Natural-language intent mapping to executable query plans | `nlp-to-query-translation` | NLQ translation flow | | Cross-model naming, granularity, and semantic consistency | `semantic-modeling-standards` | Modeling standards | ## Decision Order | Primary Need | Choose First | Reason | Escalate Next | |---|---|---|---| | Controlled definitions and shared business meaning | Business glossary | Definitions stabilize language before modeling. | Ontology | | Formal concept hierarchy and reusable meaning across systems | Ontology | Ontology encodes classes, properties, and constraints. | Semantic model | | Curated measures, dimensions, and analytic consumption | Semantic model | Semantic model optimizes reporting and metric use. | NLQ translation | | Relationship-centric exploration across many entity types | Knowledge graph | Graph design optimizes connected traversal and discovery. | Ontology | | Natural-language access to governed analytics | Semantic model | Stable metrics and dimensions ground query translation. | NLP-to-query | ## Workflow | Step | Agent Action | Test | Pass | |---|---|---|---| | 1 | Agent classifies the request by semantic layer. | Agent maps the request to glossary, ontology, model, graph, or NLQ. | Every requested concern has one primary layer. | | 2 | Agent selects the first specialist skill from the decision order. | Agent records the selected skill. | One starting skill exists. | | 3 | Agent adds dependent skills for adjacent layers only. | Agent checks handoff count. | Routing includes no unrelated skills. | | 4 | Agent aligns terminology across all chosen layers. | Agent compares canonical terms across outputs. | Core entities and measures use one vocabulary. | | 5 | Agent verifies platform fit for Fabric, Purview, and graph storage. | Agent checks integration mapping. | Each target platform has a defined role. | ## Linguistic Cross-Over Patterns | Stage | Input Artifact | Transformation | Output Artifact | Primary Skill | |---|---|---|---|---| | 1 | Raw tables, files, events | Extract candidate nouns, verbs, measures, codes | Source term inventory | `terminology-normalization` | | 2 | Source term inventory | Resolve synonyms, abbreviations, collisions | Canonical terminology set | `business-glossary-metadata` | | 3 | Canonical terminology set | Formalize concepts, relations, constraints | Ontology classes and properties | `ontology-design-patterns` | | 4 | Ontology classes and properties | Shape facts, dimensions, metrics, hierarchies | Semantic model | `semantic-layer-fabric` | | 5 | Semantic model | Map intents, entities, filters, metrics to query grammar | NLQ contract | `nlp-to-query-translation` | | 6 | NLQ contract | Execute governed query against model or graph | Natural-language query experience | `semantic-modeling-standards` | ## Microsoft Fabric Integration Guidance | Platform | Role | Integration Pattern | |---|---|---| | Fabric semantic model | Analytics serving layer | Publish curated dimensions, facts, measures, and metric definitions from governed terminology. | | Microsoft Purview | Metadata and lineage authority | Register glossary terms, data products, lineage links, and semantic ownership. | | Graph database | Relationship exploration layer | Persist high-cardinality entity relationships, cross-domain links, and traversal-heavy use cases. | | Bundle orchestration | Cross-platform alignment | Keep business terms stable in Purview, map them to Fabric model objects, and link graph entities to the same canonical identifiers. | ## Verification Matrix | Concern | Test | Pass | |---|---|---| | Terminology | Compare canonical terms against source synonyms. | One approved term exists per concept. | | Ontology | Inspect class and property reuse. | Duplicate concept structures do not exist. | | Semantic model | Inspect metric grain and dimension conformance. | Measures resolve against one declared grain. | | Knowledge graph | Inspect node and edge intent. | Every edge answers a real traversal question. | | NLQ | Trace prompt to governed metric and filter mapping. | Query translation resolves to explicit semantic objects. | | Integration | Inspect Fabric, Purview, and graph identifier mapping. | Canonical identifiers stay consistent across platforms. | ## Verification Checklist - [ ] Agent routed each concern to one primary specialist skill. - [ ] Agent applied the decision order before adding adjacent layers. - [ ] Agent aligned glossary, ontology, model, graph, and NLQ terms. - [ ] Agent defined Fabric, Purview, and graph roles. - [ ] Agent verified canonical identifier consistency. ## Common Pitfalls | Pitfall | Agent Fix | |---|---| | Start with a graph before term cleanup | Normalize terminology first. | | Build a semantic model without ontology or glossary alignment | Reconcile canonical concepts before measure design. | | Use ontology as a reporting layer | Route analytics consumption to the semantic model. | | Use Fabric model for deep relationship traversal | Route traversal-heavy use cases to the graph layer. | | Translate NLQ directly from raw schema names | Translate from governed business terms and semantic objects. |