A Mud Cake workflow can start with a spreadsheet, SQL data, another dataset, or extracted document data. Each step is durable, tenant-scoped, and recorded, so you can trace any calculated number back to its source.
The concept
Users are not simply storing datasets. They are building business processes in which datasets feed transformations, transformations produce new datasets, and those datasets feed additional transformations.
Spreadsheets
SQL data
Other datasets
Lookup + Calculate + Filter + Aggregate
Versioned, repeatable transformation steps
Calculated dataset
Materialized output with schema, storage binding, and lineage
Another workflow or downstream system
Output datasets can feed additional transformations or be consumed externally
The flow
Upload spreadsheet data, connect existing SQL tables, or extract structured information from supported business documents. Multiple ways to start, one governed platform.
Give business data a governed, inspectable home rather than leaving important logic buried in individual files and scripts. Browse, preview, inspect schema, trace lineage, and organize into folders.
Create transformations using operations such as selection, renaming, calculated columns, lookups, filters, and aggregations without writing arbitrary SQL. Lookups can join against other datasets or your existing SQL tables. The backend generates trusted SQL from validated identifiers and injects tenant predicates.
Materialize the results into datasets with schema, storage binding, and lineage. These datasets can themselves become inputs to additional workflows, so one calculation can feed the next.
Run workflows manually, on schedules from hourly to monthly, or from upstream data-refresh dependencies. A successful run durably queues downstream transforms automatically. Run history records Manual, Scheduled, or Dependency origin.
Maintain lineage, versions, execution history, tenant boundaries, and traceability throughout the workflow. Backend authorization is the authoritative security boundary, not UI filtering. Cross-tenant access is prohibited, not merely discouraged.
Transformation lifecycle
Records metadata and a Draft version. No output dataset, no rows, no lineage.
Checks the definition and runs a read-only preview, no durable output created.
Creates or updates the output dataset, schema, storage, and lineage. Does not run rows.
Materializes the published version into the output dataset and records run history.
Editing a published transformation starts a Draft revision; runs keep using the current Published version until you publish the revision. Clone copies a saved Draft or Published version into a new independent transformation.
We'll walk a real workflow through the full path, data in to chained downstream dataset.