The shape of a warehouse

Nothing clever happens here, on purpose. Staging is a faithful, typed and checked copy of what the source systems sent: dates become real dates, numbers become real numbers, obviously broken rows get flagged at the door. Daybreak's app stores order dates as text - fine for the app, poison for month-by-month analysis - and staging is where that quietly gets fixed.

• Why leaders should care: staging is your quality gate. When a source system changes something upstream, staging is where it gets caught - before it reaches a board number.
• The discipline: no business logic here. Staging makes data honest, not meaningful. Mixing the two is how untraceable numbers are born.
• What to ask your team: "What happens when a source sends garbage?" A good answer names staging checks. A worrying answer is "we usually notice."

The core is where checked parts become products: data gets modeled - deliberately arranged so business questions become easy. This is where facts and dimensions live (Part 2), where history gets kept, and where "one version of the truth" is either built or lost. It is the most engineered layer and the one your data engineers will spend most of their time on.

• One shape, every team: the core is shared. Finance and marketing read the same modeled orders - which is what makes their numbers reconcilable at all.
• The core remembers: when a customer upgrades plans, the app forgets the old plan; the core keeps it. That memory is what cohort and trend analysis stand on.

A mart is a slice of the core, pre-shaped for one audience: the finance mart has revenue with refunds already handled the finance way; the marketing mart has cohorts and channels ready to chart. Marts exist for speed and sanity - analysts should not re-derive revenue from raw parts every Monday.

• The rule that keeps marts safe: marts are built FROM the core, never from raw sources. A mart with its own private pipeline is a second version of the truth in the making.
• Marts are cheap to rebuild: if a mart is wrong, you regenerate it from the core. That disposability is a feature - the core is the asset, marts are packaging.

The app database scatters an order across half a dozen tables - great for taking orders, miserable for asking about them. The star collapses that: every business question becomes "take the fact, join the nouns you care about, summarize". Revenue by month? Fact + date. By city? Add customer. By roast? Add product. Analysts stop solving puzzles and start answering questions.

• Predictability is the gift: a new analyst can find their way around a star on day one. That is onboarding speed you can feel in salaries.
• Dashboards get faster too: BI tools are built assuming this shape. Fight the shape, fight the tools.
• The executive translation: "we model so that questions become one obvious join." That single sentence is Part 2.

If your data team debates modeling, two surnames will fly: Ralph Kimball (start with business questions, build stars, deliver value fast) and Bill Inmon (build a rigorous enterprise-wide model first, derive everything from it). Both schools are decades old and both work; most modern teams land Kimball-flavored, which is why this course teaches stars.

• You do not need to referee this debate. You need to know it is a style choice, not a correctness fight.
• The question that cuts through: "Which business question does this model make cheap?" Any modeling choice that cannot answer that is architecture for its own sake.
• Depth for the curious: your builders cover both schools properly in their session b3.

The clean division of labor in a healthy data team:

• Data engineers own staging + core: the pipelines, the checks, the modeled star. When a number is missing or late, this is the desk.
• Analysts own the marts: shaping team-specific slices and the dashboards on top. When a chart looks off, start here - it is usually the fastest fix.
• The business owns definitions: what counts as "revenue", "active", "churned" is a leadership decision wearing a technical costume. Delegating it to whoever writes the query is how metric wars start.

The question every exec eventually asks in an architecture review - and a fair one, because the slide really does show the same orders in staging, core, and a mart. The layered answer is worth rehearsing before you need it.

• They are not copies, they are stages: what the sources said, what the business means, what each team needs. A factory holds raw stock, assemblies, and packed goods too - nobody calls that waste.
• Each layer catches different failures: staging catches bad source data, core catches inconsistent logic, marts contain the blast radius of a team-specific mistake.
• Storage is the cheap part (a1's lesson): three well-governed stages cost pennies. One ungoverned pile costs executive hours every month.

Three smells that the roles have drifted, each visible from the executive floor:

• Analysts building private pipelines from raw sources - the marts have detached from the core, and divergent numbers are weeks away.
• Engineers deciding metric definitions alone - the business abdicated; expect a definition you disagree with, discovered in a board meeting.
• Every question routed to one heroic person - not a role problem, a bus-factor problem. The factory should outlive any single worker.