The Hidden Geometry of Data Models

TL;DR

Data has shape. Preserve invariants while evolving schemas.

Moves

• Normalize where correctness matters; denormalize at edges.
• Schema diffusion control: one place owns canonical facts.
• Bitemporal patterns: valid-time vs. system-time.

Example: contract test for data shape

expect(order.total).toEqual(sum(order.items.map(i => i.qty * i.price)));

Checklist

• What invariants must always hold?
• Where is data duplicated and why?
• How is history represented and queried?