Frequently Asked Questions
For enterprise Azure and Fabric data engineers
IRiS generates platform-native Lakehouse code for the Silver layer integration. The code is structured using Data Vault 2.1 methodology into Hubs, Links, and Satellites. Alongside the code, it produces a structured JSON metadata file capturing all modelling decisions, PII classifications, business definitions, entity ownership, and relationship structure. The JSON is the governance record of what was decided and why, not a by-product of code generation.
IRiS detects the load pattern for each source and generates Satellites accordingly, following CDC, delta, snapshot, and full-load precedence.
In a modern cloud data platform, schema evolution is handled in your Bronze layer. IRiS supports updating your model to accommodate new attributes through its standard development process. In most cases, Data Vault's structure means satellites can be extended to accommodate new attributes without restructuring Hubs or Links. Changes are handled with only a few minutes of development effort.
Lineage starts in the IRiS metadata itself, providing source, target, mapping, business key selection, ownership, relationship structure, and business definition for every artefact. This is the modelling-decision information that most tools never persist.
Because generated code is deployed through platform native services, the runtime data lineage is visible to lineage tools like Purview in the same way as any other platform workload.
An IRiS modelling run is metadata-only. The Assistant profiles source schemas, applies the Data Vault rules, and generates code. It does not move or transform data during modelling, so it is not a meaningful consumer of Fabric capacity.
Capacity is consumed when the generated load code runs in your environment. This is the same cost you would incur loading the Silver layer by any other means. IRiS is a development-time tool, not a continuously running production workload.
A single source table can be profiled, modelled, and have production-ready Lakehouse code generated in under 15 minutes.
The comparison point is the manual alternative: the same profiling, business key identification, Hub/Link/Satellite design, standards enforcement, and code authoring done by hand. That is a multi-hour task per source, even for an experienced modeller.