An analytics platform lead has Atlan running well. The Business Graph (its glossary) is populated, the owners are named, and the Finance terms carry a Verified certificate. A Context Engineering Studio repository is live, and an agent answers revenue questions from it. Then the AI platform team asks for something the catalog was never asked for. They want the definition of a customer, who approved it, the rule that tests it, and what it meant last quarter, in one record an agent can validate. That request is where the Atlan vs TopQuadrant comparison starts.
Atlan's answer is a glossary term, linked to a semantic model that holds the calculation, with a certificate and an activity log around both. TopQuadrant's answer is one governed asset on open W3C standards, holding the concept, the calculation, the owner, the validation rule, and every past version. This piece quotes each vendor's own pages, with dates.
Atlan is a well-documented catalog with named owners, certificates, approval workflows, a hosted MCP server, and a studio that renders context into the engines you already run. Its concept layer is glossary terms and table joins, its definition and its calculation are two objects, and its documented versioning is Draft to Active with a drift log. TQ Data Foundation holds all five things on one asset, on standards TopQuadrant helped write.
From Atlan's product documentation, because a comparison that skips its strengths helps nobody. Atlan's docs pages print no date; every one cited here was read on 2026-11-22.
| Capability | What Atlan's documentation says | Source |
|---|---|---|
| Owners and certificates on a term | "Click + under Owners to assign owners for a term." / "Choose from four certificate options - Draft, Verified, Deprecated, and No certificate." | Set up glossaries |
| Approval workflows | "Change management workflows override any permissions assigned through user roles or access policies." | Automate data governance |
| Change log | "Activity logs for metadata changes are persisted throughout the lifecycle of the Atlan instance for your organization." | Discovery FAQ |
| Hosted MCP server | "Atlan MCP is hosted by Atlan and available to all tenants, with nothing to install or run." The feature maturity matrix marks it "GA", as-of 2026-11-05. | Remote MCP overview; Feature maturity matrix |
| Renders to the engines you run | "Renders the repository as dbt semantic models in MetricFlow format" and "Exports the repository as an Apache Ossie semantic model" | Portability targets |
The three questions below are where the documentation answers differently from ours.
Atlan draws the line between a concept and a metric itself, and draws it where we do. Its "Ontology vs. Semantic Layer" page (published 2026-01-30, updated 2026-10-09) says: "A semantic layer can tell you what 'revenue' means and how to calculate it. Ask it what a customer is ... those questions fall outside what a semantic layer can answer. That's ontology territory." (source). An ontology is the formal model of what a business's concepts are and how they relate. The same page says "A business glossary is a lightweight ontology." What does Atlan ship for the concept side?
| Concept-side feature | What Atlan's documentation says | Source |
|---|---|---|
| Glossary terms | "A glossary term is the lowest unit that can exist independently inside a glossary. These terms can then be grouped into categories and linked together as related terms." | What is a glossary |
| Data models | "Atlan enables you to ingest your entity–relationship (ER) models", loaded "via an Excel template" | What are data models |
| The model an agent queries | "The semantic model: how those assets answer questions. Dimensions, facts, relationships, derived metrics, filters, and verified queries, plus the custom instructions you write to steer the agent." | What is Context Engineering Studio |
Read together: a concept in Atlan is a term with a definition, a certificate, an owner, and typed links. The model an agent queries holds tables, joins, and metrics. The pages we read describe no class with properties and constraints, held in a W3C standard, that bounds what an agent may join. A metrics layer can define a calculation. It cannot define a concept.
Credit where Atlan's argument is strong. Its Context Engineering Studio page, read 2026-11-22, says "Finance's 'revenue' and Sales' 'revenue' can both be correct." (source). The open question is whether the two revenues are related anywhere a machine can read. Our post Semantic Layer vs Ontology: Which One Says What a Customer Is makes that case in general.
In two places, by design, and Atlan documents the design plainly. Its docs say "Context Engineering Studio (CES) generates the semantic layer AI agents need to answer questions accurately." A repository "holds two things": the linked assets, "such as tables and glossary terms", and the semantic model, "the part that doesn't exist anywhere else, and the part you edit."
| Question | What Atlan's pages say | Source |
|---|---|---|
| Where does the calculation live? | Derived metrics are "Measures built on top of your facts and glossary terms" | What's in a context repository |
| What keeps the two aligned? | "linking each glossary term to the metric that computes it is what keeps the two from drifting apart" | What Is a Semantic Layer?, published 2025-12-19, updated 2026-07-30 |
| What happens when the term changes? | "The Activity tab shows what changed upstream since the model was last built, so you learn about drift from the tab rather than from a wrong answer." Then "click Regenerate to rebuild the model" | What's in a context repository |
A coherent design. The term carries the meaning and the certificate. The repository carries the math and is rebuilt when the term moves. When Finance asks which net revenue the agent used on 3 March, the answer is in two objects and a log.
Two sets of sentences, both Atlan's, both dated. The marketing page promises one thing and the documentation describes another.
| Page | Atlan's words, verbatim | Date |
|---|---|---|
| Context Engineering Studio (marketing) | "Git-like versioning. Full history. Branching. A/B testing. Staged rollouts. Rollbacks." | Undated, read 2026-11-22 |
| Build a repository (docs) | "A new repository starts in Draft. When the model looks right, activate it." | Read 2026-11-22 |
| What is Context Engineering Studio (docs) | "Redeployments replace what's live in place with no downtime." / "The Activity tab keeps the full history." | Read 2026-11-22 |
Atlan keeps history: an activity log for the life of the instance and an Activity tab per repository. What the pages we read do not describe is a rollback, a branch, a named version, or a way for a consumer to pin to the definition in force on a date. A history you can read and a version you can ask for are different things, and an audit asks for the second.
Our position, with the same weight and open to the same test. Every quoted sentence below is on a live TopQuadrant page, read 2026-11-22. TQ Data Foundation holds the definition as one governed asset. Our semantic layer use case page says: "Every metric carries its business definition, its owner, its calculation, and its mapping to physical columns. When the definition changes, it changes in one place and every consumer picks it up." The concept that metric measures is a class with properties and constraints, on RDF, OWL, and SHACL: the ground Atlan's own page calls "ontology territory".
| Same question | What TopQuadrant's live pages say | Source |
|---|---|---|
| What is a customer? | "Agents receive concepts, permitted relationships, and metric logic rather than a schema dump." / "Ontology-bounded joins. Only modelled relationships are traversable." | Agent-ready semantic layer |
| Where do the definition and the calculation live? | "Definition and calculation together. The plain-language meaning and the logic live on the same governed asset." | Semantic layer use case |
| Who owns it? | "Owner per definition. Disagreement resolves to a named person and a governed change process." | Agent-ready semantic layer |
| What did it mean last quarter? | "Version history. See what the metric meant last quarter, and what changed since." / "Reproducible. Re-run a historical question against the definitions in force that day." | Both pages |
| What leaves, and in what form? | "Everything is modelled on open W3C standards, RDF, OWL, SPARQL, SHACL and SKOS, several of which our team helped author. Your context is yours: export it and go if we stop earning the work." | TQ Data Foundation |
Atlan's own tools guide for context engineering (published 2026-05-21, updated 2026-09-21) lists TopQuadrant under "Business glossary and ontology systems". It credits "support for SKOS taxonomies, OWL ontologies, and SHACL constraint validation" and lists "SHACL validation for ontology correctness" among the pros (source). Atlan's knowledge-graph page (published 2026-07-22, updated 2026-09-21) says "A validation gate such as SHACL keeps ungoverned LLM-extracted relationships out of production graphs." (source). We agree, and every approved definition in TQ Data Foundation runs through that gate. One correction: the same tools guide marks TopQuadrant "No" under MCP Support, and our live agent-ready page says governed context is "served over API, SPARQL, and MCP for any agent framework".
Four questions, answerable from the two vendors' pages and a demo in an afternoon.
Above the catalog, as the record the catalog displays. Atlan's documented strengths are real: discovery across the estate, certificates and workflows your stewards already use, a hosted MCP server, and a studio that lands context in the engines you run. Keep them. What TopQuadrant adds is what the AI platform team asked for in the first paragraph. The concept, the calculation, the owner, the rule, and the version sit on one asset, on standards we helped write. Our post What Is Ontology Management? covers keeping that model governed. Our post What Is a Semantic Layer? says why the concept comes before the metric. Atlan answers the catalog question well. TQ Data Foundation answers what a customer is, with a name and a date on it.