Architecture Review and Design Decisions

Estimated read: 6 min

A short lecture connecting the last three back into a single set of design decisions – the checklist worth having in front of you the first few times you design a new pipeline’s layer structure, until it becomes automatic.

The full picture, end to end

flowchart TD
    Src[Source: API / DB / files] --> B[Bronze<br/>append-only, near-raw,<br/>ingestion metadata]
    B --> SV[Silver<br/>typed, validated,<br/>deduped, SCD2 where needed]
    SV -->|report-only + quarantine| SVQ[Silver quarantine]
    SV --> G1[Gold: finance revenue table]
    SV --> G2[Gold: marketing customer 360]
    SV --> G3[Gold: ops fulfillment health]

One silver layer feeding multiple, consumer-specific gold tables is the payoff of keeping the layers’ responsibilities cleanly separated – exactly what Part 2’s StepRight capstone builds, with distinct gold tables for revenue, customer 360, product performance, clickstream, and fulfillment, all from the same underlying silver.

The decisions that matter most, revisited

Decision Bronze Silver Gold
Schema rigidity Flexible (VARIANT where source is unstable) Strict, enforced Strict, consumer-shaped
Mutability Append-only Updated via MERGE, quarantine for bad rows Refreshed (table) or computed live (view)
Owns “is this correct”? No Yes Inherits silver’s correctness, adds aggregation logic
Typical consumer Data engineers debugging, audit Data engineers, analysts building gold Dashboards, executives, downstream apps

The mistake worth naming explicitly

The most common Medallion anti-pattern isn’t getting one layer’s design wrong – it’s skipping a layer under deadline pressure: a “quick” gold table built directly from bronze because silver’s validation logic isn’t ready yet. This debt compounds, because every consumer of that gold table now implicitly depends on bronze data that was never actually validated – and untangling that dependency later, once other things have been built on top of it, is far more expensive than building silver first would have been.

Where this goes next

This closes out the conceptual half of Part 1. Sections 8-10 – Lakeflow Connect, Lakeflow Declarative Pipelines, and Lakeflow Jobs – are the three pillars that actually implement this bronze/silver/gold pattern in production: ingestion, transformation, and orchestration, respectively. Everything designed conceptually in this section becomes concrete, runnable pipeline code starting with the next section’s knowledge check.


| ← Previous: Designing the Gold Layer | Next: Check Your Knowledge → | |:—|—:|


Back to top

Independent, self-authored Databricks companion notes.

This site uses Just the Docs, a documentation theme for Jekyll.