Product documentation

Datasets vs. Analysis Models

What is a Dataset?

A Dataset defines a usable SQL data boundary from one connection and selected schemas. It exposes table and field context and can carry Dataset-specific knowledge.

What is an Analysis Model?

An Analysis Model organizes reusable semantic knowledge in versions for compatible data sources and schemas. Building a version creates a candidate knowledge version; switching the active version is a separate action. Binding a model can make eligible legacy Dataset knowledge read-only.

A Dataset and an Analysis Model solve different parts of database analysis. The Dataset determines which connected source and schemas form an eligible data boundary. The Analysis Model packages reviewed semantic guidance that can be built, inspected, versioned, and activated separately. Neither object copies source database rows, replaces the source warehouse, or proves that a generated query is correct.

This separation matters because access and meaning change for different reasons. A schema owner may add or remove a field while a business owner updates the definition of an active customer. Keeping the data boundary and semantic version distinct lets teams review those changes without pretending they are one object.

Compatibility and binding do not guarantee that every question can be answered. A question may still lack the required table, relationship, definition, permission, or current source data. Visible SQL also requires review for tables, fields, joins, filters, time boundaries, aggregation, and business meaning.

Why it matters

Database analysis needs both an authorized data boundary and the context required to interpret that data. A correct boundary with weak definitions can produce the wrong business answer; strong definitions cannot supply a missing source field.

How DataFact uses them

Agents can use eligible Datasets and applicable knowledge. Analysis Model versions let teams prepare and inspect reusable guidance before selecting an active version.

Example

A revenue Dataset can expose approved schemas while an Analysis Model defines reusable entities and relationships. A reviewer still checks whether a generated query uses the revenue field and time policy intended by the question.

Limitations and edge cases

Binding can make eligible legacy Dataset knowledge read-only.
Model compatibility does not mean every table or question is supported.
Neither object validates source quality or query interpretation automatically.

Database analysis
Business Context
AI answer verification
Evaluation lifecycle