A warehouse built on proof.

01 / The context

The problem behind the project.

A practice-management database contains financial history, but its storage model is not a reporting model. The task was to make WIP, billing, allocations, receipts, and AR composable without losing their accounting meaning or lineage.

02 / The contribution

How I approached it.

1. Built ten exploratory databases, publishing the experiments and the abandoned designs, then six implementation versions frozen as correctness anchors.
2. Separated source-shaped intake from the modeled layer; enforced keys and relationships rather than admitting broken data for load speed.
3. Expressed the model through composable inline table-valued functions, with provenance columns and summary queries derived from their own detail.
4. Synchronized incrementally behind a rowversion correctness fence and reconstructed historical positions by reversing subsequent activity.

03 / The result

What the work made possible.

A delivered warehouse containing roughly 22.7 million rows of practice-management history. Its structure makes financial questions reusable and discrepancies traceable back to source records.