Session

Build the Semantic Model on Data That Doesn't Exist Yet

The report's gorgeous in dev. The visuals snap, the relationships behave, every measure totals up clean. Then it meets production volume and the slicers crawl. A relationship you trusted turns out to be many-to-many. Direct Lake quietly falls back to DirectQuery while nobody's watching. Your 5,000 polite dev rows were hiding all of it. So how do you catch that before go-live, when you can't even see production?

Because you can't just pull production down to test against. Compliance said no, and compliance is right. Masking feels like the escape hatch, but it can scramble the very cardinality and skew your model breaks on. So you end up testing against a problem you accidentally edited out. Mask the data and you mask the bug.

That's why I built Spindle.

It's an open-source, schema-aware synthetic engine, and it drops relationally correct tables straight into OneLake as Delta. You get production cardinality, real relationships, honest skew, and the dirty edge cases.

What you don't get is a single customer row or a shred of PII. 500,000 rows across 9 tables in under 3 minutes, Direct Lake ready. So you design the measures, load-test the relationships, and watch the model behave under real shape and scale, all before the first real record exists. When the production data finally lands, what surprise is left? You've already met its twin.

You'll leave with a profile-encode-generate workflow you can run against your own tenant this week, the demo's Spindle config and Fabric notebook in a public GitHub repo you can fork today, and a checklist for the model failures that tiny, clean dev data always hides.

Jonathan Stewart

Jonathan Stewart aka SQLLocks, Principal Consultant

Columbus, Ohio, United States

Actions

Please note that Sessionize is not responsible for the accuracy or validity of the data provided by speakers. If you suspect this profile to be fake or spam, please let us know.

Jump to top