Skip to main content

Glossary

Terms used throughout Atlas and this documentation.

Aggregator (vs. data provider)​

Atlas is a data aggregator and resolution platform, not a data provider: it ingests external sources — company registries (KVK, KBO/BCE, RCSL, ONRC), financial data (NorthData), sanctions/PEP/ adverse-media lists, and OSINT — and resolves them into one canonical graph with claims, survivorship, and provenance. A vendor whose product is one of those data types is a source Atlas consumes (as provenanced claims); a vendor selling a competing resolution / risk / investigation platform is a competitor; many are dual-role — a source and a competitor. See aggregator vs provider.

Atlas Console​

The React single-page frontend through which analysts run investigations and engineers configure the platform. See Frontend architecture.

Blocking​

A cheap, type-specific key that groups candidate entities so entity resolution compares an incoming record against a small candidate set instead of the whole store. See Entity resolution.

Claim​

A single assertion about an entity attribute made by one provider, carrying its own confidence and source. Multiple providers can make competing claims about the same attribute. See Claims & survivorship.

Conflict​

A recorded disagreement between providers about an attribute (within a small trust delta) or a significant change to a canonical value, raised for review rather than silently resolved. See Mutation queue.

Crew​

An LLM-driven agent (or set of agents) that performs one OSINT module, using MCP tools in an agentic loop and returning structured findings.

Data segment​

A configurable subset of configuration/scope (e.g. jurisdiction or product line) a tenant activates, managed in the Studio.

Entity​

A node in the knowledge graph — a LegalEntity, Person, or Address — defined by the ontology.

Entity resolution​

The process of deciding whether incoming data refers to an existing entity (merge) or a new one (create), using blocking and similarity scoring. See Entity resolution.

Investigation​

A unit of work that gathers, resolves, scores, and reports on a subject company. Orchestrated as a Temporal workflow.

Lineage​

Field-level provenance — the chain tracing each canonical attribute value back to the provider and claim that produced it. See Claims & survivorship and ADR-020.

Module (OSINT module)​

One of the seven focused analyses — CIR, ROA, MEBO, SPEPWS, AMLRR, DFWO, FRLS — run in parallel during an investigation. See What is Atlas.

MCP (Model Context Protocol)​

The protocol by which crews call external tools (web search, scraping, geocoding). MCP servers are registered per tenant and surfaced as LLM-callable tools.

Mutation​

A recorded change to an entity attribute — before/after values with provider attribution — forming the audit trail. See Mutation queue.

Ontology​

The versioned schema (currently v3.5) of entity types, attributes, relationships, and survivorship rules. See Ontology.

Parity​

The audit that compares the PostgreSQL system-of-record against the Neo4j projection to detect divergence. See Graph sync.

Plugin / Provider​

A data source integration declared via plugin.yaml and a mapping spec. Providers may be sync (HTTP, e.g. KVK, NorthData) or async (Temporal-orchestrated, e.g. OSINT). See Plugins.

Protected field​

An attribute (such as pep_status or sanctions_flags) that survivorship rules must never silently overwrite. See Claims & survivorship.

Provider run​

A recorded execution of a data provider (status + extracted-entity counts) used for freshness checks (stale_check) and provenance. See Ingestion pipelines.

Reference data​

Versioned datasets — FATF black/grey lists, EU high-risk third countries, secrecy jurisdictions, PEP tiers, UBO thresholds — that risk rules and matrices consult, frozen into snapshots at publish time so historical scores stay stable. See Reference data.

Risk indicator​

A finding emitted during an investigation carrying a severity (critical/high/medium/ low) and one of the eleven risk categories. See Risk scoring.

RLS (Row-Level Security)​

The PostgreSQL mechanism that scopes every query to the current tenant. The backbone of Atlas multi-tenancy. See Security & multi-tenancy.

Risk matrix​

An operator-configurable grid that maps entity characteristics (jurisdiction, industry, UBO clarity, …) to a risk level and score. See Risk scoring.

Source contract​

A declarative description of which ontology fields a provider is allowed to emit, parallel to its mapping spec. See Source contracts.

Studio​

The no-code area of the Atlas Console where compliance engineers configure ontology mappings, risk matrices, data segments, providers, reference data, and crew/agent/prompt/model settings without code. See Extending Atlas.

Survivorship​

The rules that select the winning claim per attribute, weighted by provider trust and guarded by protected fields. See Claims & survivorship.

Temporal​

The durable workflow engine that orchestrates long-running investigations so they survive restarts and are fully observable.

Tenant​

An isolated customer environment. Each tenant has its own Keycloak realm, data segment, and provider credentials; isolation is enforced by RLS.

TranslationRegistry​

The boot-time index (Phase 114) that derives canonical field paths from plugin specs and the active ontology, validating that plugins target a compatible schema version.

UBO (Ultimate Beneficial Owner)​

The natural person(s) who ultimately own or control a company, computed by walking ownership chains in the graph.

Workflow engine​

The low-code, feature-flagged subsystem for authoring multi-step compliance workflows (phases, gates, decision options, escalation, a task inbox, portal forms). Its lifecycle capabilities are on the Product roadmap.