01 / The problem
A quality score tells you something is wrong.
Lineage helps you find where.
Data moves through systems, transformations, and teams before it reaches the people who use it. When quality breaks down, a score alone cannot explain where the issue began or which downstream data may be affected.
Data stewards responsible for operational governance needed to see quality in the context of those relationships. Observability brought lineage and quality into the same view so they could follow the data, locate potential breakdowns, and decide where to investigate.
I owned the experience from zero to one, working with product management and engineering in a close triad. We used a Build, Measure, Learn approach: develop a direction, test it early, and use what we learned to shape the next iteration.
02 / The relationships
Make the relationships
readable.
I started with a compact node graph I had designed for our data visualization library. It could represent many connected objects in little space and worked well for tracing a specific path. But at an estate level, the meaning of those connections was harder to understand.
The experience needed to communicate two relationships: containment, showing which assets belonged to a data product or organizational group, and flow, showing how data moved between them.
What belongs together?
Nested cards show assets within products and groups within a hierarchy.
Where does data move?
Directional connections show movement between assets and products.
Cards made room for names, metadata, and quality indicators. Nesting made ownership and hierarchy explicit. The tradeoff was space: each object became more useful, but the same estate now needed a much larger canvas.
03 / Designing for scale
Real data changed
the design.
Engineering prototyped the card view with actual data, exposing a problem that smaller examples did not: the experience did not scale to larger customers.
I explored deeper nesting and smaller, collapsible cards to keep more of the estate manageable. Progressive disclosure let users choose where to reveal detail. Icons and labels also reduced reliance on color alone to communicate quality.
As we tested larger estates, the team also anticipated performance problems with an estate-wide view. The answer required changing the scope of the experience as well as its presentation.
Scoping the canvas to a governance domain reduced the amount of data it needed to display. Incoming and outgoing lineage could still surface relevant objects from outside that domain, preserving the context needed to follow a quality issue.
04 / Progressive disclosure
Let users choose
where the detail lives.
Collapsed cards provide a summary for scanning. Expanding a card reveals its contents. Cards can be resized, with internal scrolling to keep a long list of assets from taking over the canvas.
Scoping, filtering, and following lineage solve different parts of the same problem. The domain sets a useful boundary, quality filters narrow attention, and connections help stewards trace the issue across that boundary.
05 / Into preview
A focused view
with room to go deeper.
Observability shipped in preview, giving data stewards a way to inspect lineage and quality within a selected governance domain. Domain-level scoping mitigated the performance concerns that emerged as the team explored larger estates.
The largest domains could still produce dense views. Further scoping remained the next design challenge. The work established a practical way to navigate that complexity: begin with a manageable context, preserve meaningful connections, and reveal detail where the investigation takes you.