DBRS Tagging Specification v1.0 — DBRS-ID: dbrs_75fe51b2 canonical

Summary-of-Content

The DBRS Tagging Specification v1.0 establishes document tagging rules for deterministic discoverability and strict LLM grounding by separating verbatim and conceptual tags.

DBRS Semantic Core

Resolver Text

DBRS Tagging Specification v1.0 Status: Stable Scope: DBRS Frontmatter, Indexing, UX & LLM Consumption Applies to: Websites, Knowledge Hubs, Intranets, DBRS Repositories 1. Purpose The DBRS Tagging Specification defines how tags are used to ensure: deterministic discoverability strict non-hallucinating LLM behavior high-quality user experience (UX) verifiable and citable references This specification intentionally separates discoverability from interpretation. 2. Core Principle Meta-tags may explain meaning. Textual tags must enable finding. DBRS does not infer relevance through semantic guessing. All primary tags must be explicitly grounded in the source text. 3. Tag Categories 3.1 Textual Tags ( tags ) Definition: Textual tags are literal identifiers used for search, navigation, UX, and strict LLM grounding. Rules (hard): Each tag MUST appear verbatim in the document Allowed locations: title headings (h1–h3) body text No normalization No synonym generation Case-sensitive Punctuation-sensitive Purpose: User search & navigation Deterministic LLM entry points Prevention of false negatives Trust-preserving UX Example: tags: - IT Service+ - Managed IT - Bestands-IT 3.2 Meta Tags ( tags_meta ) Definition: Meta tags describe conceptual, thematic, or analytical aspects that may not be explicitly named in the text. Rules: MAY be abstracted MAY use normalized or English terms MUST NOT replace textual tags MUST NOT be required for basic discoverability Purpose: Context enrichment Clustering and classification Relevance analysis DBRS internal reasoning layers (QML, CCR, CCA) Example: tags_meta: - existing systems - legacy systems - ongoing support - security risks 4. Mandatory Separation Textual tags and meta tags MUST be stored in separate fields. Mixing both categories in a single tags field is explicitly forbidden. Invalid example (DO NOT USE): tags: - IT Service+ - legacy systems - security risks 5. Validation Rules 5.1 Textual Tag Validation (Mandatory) For every entry in tags: The tag MUST appear verbatim in the document text Otherwise: The build MUST fail The DBRS index entry MUST NOT be published This rule ensures DBRS remains deterministic, verifiable, and citation-safe. 6. Impact on UX and LLM Behavior Without this specification Valid content may not be found Strict prompts cause false negatives Users experience inconsistent results LL Ms appear unreliable With this specification Search results are predictable Strict prompts remain intact LL Ms answer with confidence and precision User trust increases 7. Design Rationale DBRS does not aim to “understand everything”. DBRS aims to reference correctly. Meaning may emerge. Addressability must be explicit. This distinction is fundamental to DBRS. 8. Versioning Version: 1.0 This specification is considered stable Future versions may extend validation tooling but will not weaken rules 9. Summary tags are literal, textual, and UX-critical tags_meta are conceptual and analytical Both are required Both serve different purposes DBRS enforces their separation deliberately End of DBRS Tagging Specification v1.0

CCR Context

Links

DBRS Semantic Field Guide