Skip to content

Glossary

Business Value

A shared vocabulary bridges the gap between business stakeholders and engineering teams. This glossary provides dual definitions — functional (for executives and product teams) and technical (for engineers and SREs) — ensuring clear communication across all audiences and reducing misunderstandings during planning, incident response, and product discussions.

This glossary defines key terms used throughout the Labeeb platform documentation. Each term includes both a functional definition (for executives and product teams) and a technical definition (for engineers and SREs).

A

Adapter
Business: A connector that enables the platform to ingest content from a new source type without changing the core system.
Technical: An implementation of the IngestionAdapter interface that normalizes source-specific payloads (RSS, HTML, API) into a standard FactCheckReviewDTO for pipeline processing.
Annotation
Business: A label or judgment attached to content — either generated by AI or added by an analyst — that enriches the data with context.
Technical: A polymorphic model linking machine or human labels (stance, category, quality score) to articles, claims, or entities via annotatable_type and annotatable_id.
Article
Business: A piece of news content — the primary unit of data that flows through the entire platform.
Technical: The core database model representing normalized content with source metadata, language, publication date, and processing status. Stored in PostgreSQL and indexed in OpenSearch.

B

BM25
Business: A fast keyword-matching search method that finds documents containing specific words, ideal for precise queries.
Technical: Best Matching 25 — a probabilistic information retrieval algorithm that scores documents based on term frequency, inverse document frequency, and document length normalization. Used as one leg of the hybrid search strategy.

C

Check-Worthiness (S1)
Business: The platform's ability to automatically detect which statements in a news article are factual claims worth verifying — replacing manual triage.
Technical: The S1 pipeline stage that uses zero-shot Natural Language Inference (NLI) via a three-tier provider chain (NLP Lab → Cloudflare → Heuristic) to classify sentences as check-worthy or not, returning confidence scores.
Claim
Business: A factual statement extracted from content that the platform verifies against evidence.
Technical: A database model linked to an article, containing the extracted text, check-worthiness score, stance verdict, and associated evidence references.

D

DFD (Data Flow Diagram)
Business: A visual map showing how data moves through the platform — from ingestion to user-facing results.
Technical: A Mermaid-rendered diagram documenting the flow of data between services, stores, and external actors. Used in architecture pages to represent system topology.
DTO (Data Transfer Object)
Business: A standardized data package that ensures content is consistently structured as it moves between services.
Technical: A PHP or Python class that defines the shape of data passed between layers (e.g., FactCheckReviewDTO, ClaimDTO). Enforces validation and type safety at service boundaries.

E

Entity
Business: A person, organization, or place detected in content — used to connect related articles and build knowledge graphs.
Technical: A named entity extracted by the S2 NER pipeline (via NLP Lab or SinaTools), stored with type classification, span positions, and article linkage.
Evidence
Business: Supporting or contradicting information retrieved from indexed sources that backs up or challenges a claim.
Technical: A ranked set of documents returned by the hybrid retrieval pipeline (BM25 + kNN + RRF fusion), hydrated into retrieve_pack objects with source metadata, relevance scores, and diagnostic information.

G

Gatekeeper
Business: A quality filter that prevents low-quality or duplicate content from entering the processing pipeline, protecting output integrity.
Technical: A pre-ingestion assessment stage that evaluates article quality, language confidence, and deduplication signals before allowing content to proceed to the full pipeline.

H

Horizon
Business: The platform's background job processing system that handles thousands of tasks concurrently without blocking user-facing operations.
Technical: Laravel Horizon — a Redis-powered queue dashboard and supervisor for managing queue workers, job retries, and throughput monitoring.
Hybrid Search
Business: A search approach that combines keyword precision with semantic understanding, delivering more relevant results than either method alone.
Technical: A dual-leg retrieval strategy executing BM25 (keyword) and kNN (vector/semantic) searches in parallel, then fusing results using Reciprocal Rank Fusion (RRF) for a combined relevance ranking.

I

IaC (Infrastructure as Code)
Business: Automated, version-controlled infrastructure provisioning — enabling rapid, repeatable, and auditable deployments.
Technical: Terraform configurations in terraform/ that define DigitalOcean, Cloudflare, and monitoring resources as declarative code.
IndexLane
Business: A dedicated processing path for indexing content into the search engine, ensuring search results stay current.
Technical: A queue-based subsystem in the ingestion pipeline that handles OpenSearch document indexing, update, and deletion operations with retry logic and batch optimization.
ISM (Index State Management)
Business: Automated lifecycle management for search data — old data is archived or deleted automatically to control costs.
Technical: OpenSearch Index State Management policies that automate index rollover, retention, and deletion based on age or size thresholds.

K

kNN (k-Nearest Neighbors)
Business: A semantic search method that finds documents with similar meaning, even if they use different words.
Technical: k-Nearest Neighbors vector search using dense embeddings. Documents and queries are represented as high-dimensional vectors; similarity is computed via cosine distance in OpenSearch's vector engine.

N

NER (Named Entity Recognition)
Business: The platform's ability to automatically identify people, organizations, and locations mentioned in content.
Technical: The S2 Information Extraction pipeline stage that uses NLP Lab or SinaTools models to detect and classify named entity spans within article text.
NLI (Natural Language Inference)
Business: An AI technique that determines whether one statement logically supports, contradicts, or is unrelated to another.
Technical: A zero-shot classification approach used in S1 check-worthiness detection, where a premise-hypothesis pair is evaluated for entailment, contradiction, or neutrality.

O

Orchestrator
Business: The intelligent routing layer that decides how to process each request — balancing speed, cost, and quality automatically.
Technical: The AI-Box V2 orchestrator that implements a routing engine with policy guards, complexity-based decision logic, and dual-path streaming (Edge SSE push + fallback SSE).

R

RRF (Reciprocal Rank Fusion)
Business: A ranking method that intelligently combines results from multiple search strategies to surface the most relevant content.
Technical: A rank aggregation algorithm that merges ranked lists from BM25 and kNN retrieval legs using the formula 1/(k + rank). Implemented natively in OpenSearch with a Python fallback for compatibility.

S

S1 (Stage 1)
Business: The first AI analysis stage — identifies which statements in an article are factual claims worth checking.
Technical: The check-worthiness classification stage in the AI-Box pipeline, using a provider chain (NLP Lab → Cloudflare → Heuristic) with Top-K optimization and confidence thresholds.
S2 (Stage 2)
Business: The second AI analysis stage — extracts structured information (people, places, organizations) from content.
Technical: The information extraction stage using Named Entity Recognition (NER) via NLP Lab, with segment filtering, profile selection, and span building for entity resolution.
SLI (Service Level Indicator)
Business: A measurable signal that tells you how well a service is performing right now.
Technical: A quantitative metric (e.g., request latency, error rate, throughput) used to evaluate service health against defined thresholds.
SLO (Service Level Objective)
Business: A target for service reliability — the promise of how available and fast the platform will be.
Technical: A defined target range for an SLI (e.g., "P95 latency < 500ms", "99.5% availability"), used as the threshold for alerting and capacity planning.
SSE (Server-Sent Events)
Business: A streaming technology that delivers real-time results to users as they are generated, rather than making them wait for complete processing.
Technical: A unidirectional HTTP streaming protocol used by the orchestrator's dual-path streaming system (Edge push + SSE fallback) to deliver incremental fact-check results to the frontend.
Stance Detection
Business: The platform's ability to determine whether a piece of evidence supports, refutes, or is neutral toward a claim.
Technical: An AI classification stage that evaluates evidence-claim pairs using timeout-budgeted inference, decision gates, and implicit contradiction detection to assign stance labels (support/refute/neutral).

V

Vectorize
Business: Cloudflare's global vector database that enables lightning-fast semantic search from 300+ edge locations worldwide.
Technical: Cloudflare Vectorize V2 — a managed vector index storing document embeddings for kNN similarity search, integrated via the UpsertToVectorize pipeline stage.
Verdict
Business: The platform's final assessment of a claim — combining evidence, stance analysis, and confidence scoring into a transparent fact-check result.
Technical: An aggregated output of the full pipeline (S1 → S2 → Evidence Retrieval → Stance Detection), stored as a structured record linking claims to evidence with stance labels and confidence scores.

Using This Glossary

When writing or reviewing documentation, use these terms consistently. Business definitions should appear in executive-facing pages and the Business Intelligence section. Technical definitions belong in architecture, runbook, and API reference pages.