Glossary¶

Technical terms used throughout dbt-nova documentation.

Search & Indexing¶

Term	Definition
BM25	Best Match 25 - a text relevance scoring algorithm that considers term frequency and document length
RRF	Reciprocal Rank Fusion - combines multiple ranking signals by summing reciprocal ranks
SPLADE	Sparse Lexical and Expansion - learned sparse vectors that capture semantic meaning
ANN	Approximate Nearest Neighbors - fast vector search using locality-sensitive hashing
Quantization	Reducing vector precision (e.g., 32-bit → 8-bit) for faster search at minimal accuracy cost
Tantivy	The Rust search engine library powering dbt-nova's lexical search
Reranker	Cross-encoder model that re-scores top candidates for improved relevance
N-gram	Substring indexing (e.g., 3-grams: "abc", "bcd") for partial matching

Term	Definition
unique_id	Full identifier for a dbt entity: `resource_type.package.name` (e.g., `model.jaffle_shop.orders`)
relation_name	Physical database location: `database.schema.table`
manifest.json	dbt artifact containing all project metadata, generated by `dbt compile` or `dbt build`
lineage	The dependency graph showing how data flows between models and sources
resource_type	Entity category: `model`, `source`, `test`, `seed`, `snapshot`, `macro`, `exposure`, `metric`

Term	Definition
Nova meta	Custom metadata fields under `meta.nova.*` for enhanced search and governance
Persona	Search profile (`analyst`, `engineer`, `governance`) that adjusts ranking weights
detail	Controls response detail: `standard` (summary) or `full` (complete entity)
Metadata score	Quality score (0-100) measuring documentation, tests, and governance compliance

Term	Definition
Storage root	Base directory for indexes and caches (default: `.dbt-nova/`)
Instance	Isolated index build, allowing multiple manifests or versions
Circuit breaker	Protection mechanism that temporarily disables failing operations