Use this glossary for the data, processing, search, model, evaluation, and operating terms used throughout CoreCortex.
Corpus and source data
Term
Meaning
Corpus
A corpus is a collection of documents, records, audio, images, or other information about a subject. CoreCortex prepares it for search, keeps results linked to their sources, and tests retrieval with reviewed questions.
Dataset
One coherent part of a corpus, such as a document archive, database export, or recording series
Source
The system, file collection, feed, or export from which a dataset was obtained
Snapshot
The exact source contents processed in one reproducible run
Modality
The form of source information: text, structured data, audio, video, image, or another media type
Source identity
The stable identifier that connects a CoreCortex record to the item in its source
Record
The normalized representation of one source item inside a dataset
Attribute
A typed field on a record used for meaning, filtering, policy, or display
Representations and derived data
Term
Meaning
Normalized record
A record in the stable CoreCortex shape, with identity, content, attributes, access, and source details
Fragment
A retrievable unit derived from a record; the documentation also calls this a passage when it contains text
Chunk
The source span or content selection used to create a fragment
Processor
Versioned code that converts, enriches, chunks, embeds, or extracts information
Provenance
The source, processor, settings, and timestamps that explain how an object was produced
Relation
A typed, directed connection between records or fragments, with its own source evidence
Access policy
The principals or rules that determine who may retrieve a record and its derived data
Canonical data
Records and reviewed judgments kept independently from any search or graph backend
Search and storage
Term
Meaning
Schema
The fields, types, identities, and constraints expected at a particular boundary
Field role
How a field is used in search: lexical text, vector, exact filter, relation, or returned content
Index
Selected corpus data arranged for fast lookup or ranking
Projection
A rebuildable copy of canonical data in a search, relational, vector, or graph backend
Backend
The storage or search system that holds a projection
Adapter
The CoreCortex implementation that translates a common operation to a backend
Embedding
A numeric representation used to compare semantic similarity
Lexical search
Search that ranks words, terms, identifiers, and phrases, commonly with BM25
Queries and retrieval
Term
Meaning
Query
The question or lookup request sent to retrieval
Retrieval plan
The named configuration that chooses indexes, candidate methods, filters, ranking, and result limits
Candidate
A record or fragment returned by one first-stage search method before final ranking
Filter
An exact condition that removes results outside scope, time, category, tenant, or access policy
Ranking
The ordering of matching candidates by a score or policy
Fusion
A method that combines candidate lists; reciprocal rank fusion, or RRF, combines their rank positions
Top K
The number of highest-ranked results retained or sent to the next step
Reranker
A second-stage scorer that reorders a smaller candidate set
Graph-assisted retrieval
A bounded traversal of evidenced relations followed by retrieval of the linked source content
Trace
The recorded plans, index versions, scores, filters, and identities used for one request
Models and processors
Term
Meaning
Strategy
A registered implementation selected through a stable CoreCortex interface
Chunking strategy
The rules that create fragments from normalized record content
Embedding provider
An implementation that creates vectors and records the model identity and revision
Language-model provider
An implementation that sends authorized evidence to a selected language model
Relation extractor
An implementation that produces evidenced triples or typed relations
Response plan
The configuration for evidence limits, model invocation, citation requirements, and abstention
Synthesis
The optional step that uses retrieved evidence to produce an answer or other model output
Evidence envelope
The authorized fragments supplied to a model for one response
Citation
A reference from generated output to a fragment inside its evidence envelope
Abstention
An explicit response stating that the available evidence cannot support an answer
Evaluation
Term
Meaning
Reviewed judgment
A question plus the records or fragments that a subject-matter reviewer accepts as relevant
Golden data
The versioned set of reviewed judgments used to compare configurations
Hard negative
A plausible result that a reviewer has marked irrelevant for a specific question
Expected-empty query
A question designed to return no supported result
Retrieval evaluation
Measurement of whether reviewed evidence appears and where it ranks
Generation evaluation
Measurement of required facts, citation support, faithfulness, and abstention
Hit@K
The share of questions with an accepted result among the first K results
MRR
Mean reciprocal rank; a score that rewards placing the first accepted result near the top
NDCG
Normalized discounted cumulative gain; a rank score that can use graded relevance
Faithfulness
Whether generated claims are supported by the supplied evidence
Miss
A query or answer that fails a reviewed expectation and needs analysis
Runs and promotion
Term
Meaning
Configuration
The declared datasets, processing steps, storage, indexes, retrieval plans, and optional response plans
Configuration revision
An immutable saved version of a configuration
Configuration fingerprint
A SHA-256 identifier for the exact settings used in a run
Ingestion batch
A validated set of records and operations prepared for one apply step
Receipt
Machine-readable evidence that a batch or projection operation ran against named inputs and settings
A–E retrieval experiments
Ordered comparisons of representation, chunking, embeddings, retrieval, and full evaluation
Gate
A required quality, policy, integrity, latency, or cost condition
Certification manifest
The final status and evidence that show whether a selected corpus configuration is ready for pipelines
Read the detailed guides
The glossary defines the terms. The guides below explain how to configure and use them.