Data Engineering · Architecture tutorial
Medallion data pipeline diagram: Bronze, Silver and Gold
Build an editable medallion data pipeline diagram. Explain ingestion, Bronze/Silver/Gold layers, orchestration, quality gates and recovery decisions.
By DrawArc · Updated November 17, 2026 · Practical walkthrough
Open & edit this diagram — no login requiredEdits use a separate browser-local template draft. Existing saved files stay untouched.
Logical reference, not a verified deployment. Sources, assumptions & corrections
Social sharing is available for public drawarc.io template and read-only links. You can select the local preview link above to copy it manually.
What the three layers communicate
A medallion architecture organizes data by refinement level. Bronze holds ingested source data, Silver contains validated and normalized records, and Gold provides models prepared for downstream business use. These labels describe data responsibilities; they do not require exactly three tables or three running services.
The DrawArc starter connects operational sources, ingestion jobs, raw storage, Spark transformations, validated data, business models and BI consumers. Airflow appears as an orchestration dependency. Start by separating data movement from scheduling so readers can distinguish what processes records from what coordinates the work.
Draw the data path and the quality gates
Follow one source change from ingestion to a dashboard. Label whether ingestion uses an extract or change data capture, then describe what is retained for replay. On the transformation boundary, record the checks that decide whether records proceed or require investigation. Name the owner of rejected data instead of drawing an unexplained failure bucket.
The starter's Gold component represents business modeling. Customize it with an actual output such as daily order totals or a consumption-ready dataset. Keep definitions and freshness expectations next to the output so consumers know what the diagram promises. A schedule succeeding does not, by itself, establish that the business result is correct.
- Distinguish data arrows from orchestration arrows.
- Name duplicate, schema and missing-field handling.
- Assign a data-quality owner and an expected freshness window.
Make recovery and access part of the design
Record what happens when the source is late, a transformation fails or a quality check rejects a batch. Show the boundary at which processing can safely restart, and document how a replay avoids producing duplicate business results. Identify which downstream outputs should pause when upstream validation fails.
Raw records can contain more sensitive information than a curated model. Define access and retention for each layer rather than granting every dashboard consumer access to raw input. Draw lineage and deletion responsibilities where they affect the workflow. The reference does not configure storage permissions, retention or a recovery mechanism; those remain implementation decisions.
Choose the smallest useful pipeline
A layered architecture can make responsibilities clear, but it also adds transformation, storage and operational work. Keep a layer only when its contract serves the workload. Consider how incremental runs, historical backfills and changing business definitions affect the design, and write down which trade-off led to the chosen structure.
For a worked exercise, map one daily sales report. Name the source, define an acceptable delay and list the checks needed before business users read the Gold output. Add a recovery note for an incomplete source extract. This is an illustrative reference; it is not a customer result or a certified production architecture.
Try it yourself
- Open the lakehouse starter and replace generic names with one real dataset.
- Label a quality gate and an operator response when it fails.
- Show the replay boundary and how duplicate output is prevented.
- Review access and retention for raw versus curated data before sharing.
Further reading
Consult the platform documentation when implementing and validating your design.
More architecture tutorials
AWS
AWS 3-tier architecture explained
Follow a request through CloudFront, load balancing, application services and PostgreSQL. Learn tier boundaries, failure handling and security, then edit the diagram.
AI / LLM
Production RAG architecture explained
Understand ingestion, authorized retrieval, grounded generation and evaluation in a production RAG system. Customize an editable retrieval-augmented generation diagram.
Microservices
Kafka event-driven architecture explained
Learn Kafka topics, consumer groups, ordering, retries and database consistency through an order workflow. Open and customize the event-driven architecture diagram.
Kubernetes
Kubernetes architecture diagram explained
Follow a request through an editable Kubernetes architecture diagram. Explain ingress, Services, workloads, identity and failure paths using an EKS starter.