Ontology
Tattva's model of what your data means, and why it lets you trust the numbers in an answer.
The ontology is Tattva's model of your business: which metrics exist and exactly how each is calculated, what your columns mean, how tables join, and which conventions apply (financial year, currency, what counts as a new customer). It is what turns "a model reading a spreadsheet" into answers you can check.
Your data analysts build and look after it. You benefit from it every time you ask a question.
What is in it
- Metrics, each with a stored definition: the exact SQL or formula your data analysts approved, and the inputs it needs, such as the period.
- Metric families: one business word that covers several variants. "Revenue" might mean gross, net of returns, or net of returns and discounts. The family says what to do when you ask for "revenue": show every variant, ask you which one, use a default and mention the others with their figures, or use a default and not mention the others.
- Concepts and conventions: the business terms and rules your team uses. Ask "What does RTO mean?" and, when a concept (or a metric family or metric) has that name, the answer is its description exactly as your team wrote it. See Answer scope.
- Column meanings: what each column holds, its role, and how it may be added up.
- Relationships: how tables safely join.
Make, Manage, Maintain
Sidebar → Ontology. It opens on the Knowledge Graph. The page has three modes:
- Make (what exists): the definitions themselves. Here you can look up any metric. Workspace Library holds the definitions shared by every project.
- Manage (how it connects): the Knowledge Graph of sources, concepts, metrics and relationships, and References.
- Maintain (what needs fixing): Health, Conflicts, the Review queue of proposed definitions and correction cards, Feedback (what people said about answers), Promotion requests and Versions.
In plain words: data analysts make the definitions, manage how data connects, and maintain it as sources change. Most of that work happens in Make and Maintain; see For data analysts.
Look up how a metric is defined
- Ontology → Make → Data. The Domain sub-tab shows the Domain Glossary.
- Find the metric. Its row shows what kind of definition it has and whether that definition passed its checks.
- Click Definition →. The Definition tab shows the SQL, the Parameters (where each input comes from), the Validation results and the Source it was written from.
- Under Try it, pick dates and click Run to compute the metric yourself.
Metrics without a stored query show Details → instead.
Why you can trust the numbers
- Code computes defined metrics. When your question uses a metric with a stored definition, Tattva runs that definition in code, with your period and filters, and code lays out the answer's tables and chart from the result. A model may add a few sentences about the figures; it never types one.
- Definitions are checked before they are used. A data analyst reviews each new definition, and Tattva runs it against the warehouse. Definitions that fail the basic checks cannot be accepted.
- Every answer shows its work. On the Trail tab, the Analyze card has a Compute block per metric: the value, the filters, where each input came from, and the SQL that ran.
- What a model writes is checked. A second model lists every claim the prose makes, and code checks each against the figures. A sentence that fails is revised once, then dropped whole.
- Unsettled data is flagged. Some metrics, such as returns, take days to settle. Figures that include unsettled days are marked provisional.
- Nothing else is computed. Tattva runs no queries of its own: a figure no stored definition gives is not shown, and the answer says which part is missing.
See How answers are computed for the full detail.
When your data changes
When sources change, data analysts rebuild the ontology from Make. Until they do, the Make page shows "Sources changed since last build". Column meanings edited since the last build also wait for the next build before answers use them.
Every change is recorded under Maintain → Versions, and can be reverted.
What you might do as a business user
- Look up a definition before you quote a number outside the team.
- Answer a clarification. If a metric family is set to "ask which one", Tattva asks you which variant you mean.
- Tell your data analyst when a definition does not match how the business works. They change it; you don't need to.
Related
- For data analysts: building and maintaining the ontology
- Workspace library: definitions shared across projects
- Review an answer