Monorepo surgery, CI discipline, semantic layers, and cost levers — Discord's petabyte dbt to Datadog's 5,000-person self-serve.
Past a few thousand models, dbt stops being a tool and becomes a platform you engineer: Discord's petabyte-scale surgery on dbt internals, Checkout.com's 27-project CI library, the Airbnb/DoorDash/Netflix semantic-layer canon, and the SQLMesh-era platform decisions.
Why certified tables don't produce consistent numbers: define-once platforms, the semantic compile path with typed entities, the six-type metric taxonomy, and consolidation as the other road.
The CEO asks the simplest question in the business: *which city had the most bookings last week?* Data Science answers. Finance answers. The numbers differ by eleven percent. Both teams are competent; both queries are defensible; each sits on its own carefully-built derived table, forked from the same certified core data two years ago and drifted since. The meeting that follows isn't about bookings anymore — it's about whether *any* number in the company can be trusted. Everyone agrees "we should define metrics once." Nobody agrees on what that actually requires. What does it?
The full week 3 brief is part of LeetData Pro.
