Further Reading: Data Modeling

Sources are tagged Tier 1 (confident it exists, recommended without reservation) or Tier 2 (real and worth seeking, but confirm the current edition, version, or URL yourself).

This is the chapter with the best literature behind it. Dimensional modeling was written down thoroughly, by people who cared about being understood, thirty years ago, and almost none of it has gone stale.

The canon

Ralph Kimball and Margy Ross, The Data Warehouse Toolkit: The Definitive Guide to Dimensional Modeling, 3rd edition (Wiley, 2013). The book this chapter compresses. Chapter 1 gives the four steps and the case for dimensional modeling; Chapter 2 covers the bus matrix; Chapter 3 works through retail sales in detail and is the closest published analogue to Kestrel. The remaining fifteen chapters are worked case studies by industry — inventory, procurement, customer relationship management, healthcare, education — and the right way to use them is as a reference when you meet that domain, not as a cover-to-cover read.

Read Chapters 1–3 before Chapter 19 of this book. They are about 120 pages and they will make the dbt chapter substantially easier. Tier 1.

Ralph Kimball and Margy Ross, The Kimball Group Reader, 2nd edition (Wiley, 2016). Collected short articles, many of them originally magazine columns, and the format suits the material — each is a self-contained answer to one modeling question. The articles on grain, on slowly changing dimensions, and on "the ten essential rules of dimensional modeling" are the ones worth seeking out first. Tier 1.
Bill Inmon, Building the Data Warehouse, 4th edition (Wiley, 2005). The other side of the argument in §6.2's 📐 callout. Read at least the first two chapters, because understanding the normalized-enterprise-warehouse position properly is what lets you recognize when you are in a situation that calls for it — a large, regulated, slow-moving organization with many source systems modeling the same entities. Tier 1.

On slowly changing dimensions specifically

Kimball's original articles on SCD types, collected in the Reader and widely summarized online. The types beyond 2 — particularly Type 4 mini-dimensions and Type 6 hybrids — are worth knowing exist even though you will rarely build them, because recognizing that you need one saves you from inventing a worse version. Tier 1 in the book; Tier 2 for online summaries, which vary in quality and occasionally get Type 6 wrong.
The dbt snapshots documentation (docs.getdbt.com, "Snapshots"). dbt implements Type 2 dimensions as a first-class feature, and reading how it does so — the check and timestamp strategies, dbt_valid_from/dbt_valid_to, and the invalidation of hard deletes — is a concrete grounding for §6.5. Chapter 20 of this book uses it. Tier 2 — versioned documentation.

On the modern context

Maxime Beauchemin, "Functional Data Engineering" (2018). Recommended in Chapter 2 and again here for a different reason: its argument for immutable, reproducible, partition-based pipelines interacts directly with dimensional modeling, and it is where the tension between "rebuild everything deterministically" and "maintain Type 2 history incrementally" is best articulated. Chapter 20 lives in that tension. Tier 2.
The dbt community's writing on "one big table" versus star schemas. A genuine, current disagreement: some practitioners argue that with columnar storage and cheap compute, a single wide denormalized table per business process is simpler and fast enough, and that dimensions are a relic of join-expensive hardware. The argument has real force for small analytical surfaces and breaks down on conformed dimensions and on Type 2 history. Worth reading both sides; search for "one big table dbt" and read the discourse rather than any single post. Tier 2 — forum and blog content, contested and moving.
Joe Reis and Matt Housley, Fundamentals of Data Engineering, Chapter 8 ("Queries, Modeling, and Transformation"). A shorter treatment of modeling than Kimball, positioned within the lifecycle, and useful specifically for its coverage of Data Vault — a third modeling approach this chapter does not cover, which has a real constituency in regulated industries and which you should be able to recognize. Tier 1.

On Data Vault, since this chapter skipped it

Daniel Linstedt and Michael Olschimke, Building a Scalable Data Warehouse with Data Vault 2.0 (Morgan Kaufmann, 2015). Data Vault models hubs (business keys), links (relationships), and satellites (descriptive attributes with history), which produces an extremely auditable and extremely join-heavy structure. It is genuinely well suited to environments where full auditability and source traceability are regulatory requirements, and it is genuinely painful to query directly — which is why Data Vault implementations almost always have a dimensional layer on top for consumption. Know it exists; adopt it only if you can name the regulatory requirement driving it. Tier 2 — a specialist book; confirm the edition.

On the organizational half of the problem

Chapter 1's Case Study 1 of this book, and the writing on metric layers referenced in Chapter 2's further reading. §6.8's five resolved ambiguities are the modeling side of the same problem — a metric definition and a fact table's grain are two views of one decision, and the reason definitional work is hard is not technical. Tier 1 (it is in your hand).
Cassie Kozyrkov's writing on decision-making and metrics. Not a data engineering source, and the most useful available treatment of why organizations resist committing to a single metric definition — because a definition forecloses interpretations that people are relying on. Read it before the meeting where you present nine spellings of "churn." Tier 2 — blog and talks.

If you only read one thing

Read Kimball and Ross, Chapters 1 through 3. It is about 120 pages, it is the source of everything in this chapter, and it is written unusually clearly for a technical book of its era. Chapter 3's retail sales example is close enough to Kestrel that you can read it as a worked solution to this chapter's exercises.

If you have one hour rather than one weekend, read Chapter 1 alone — the four steps and the argument for declaring grain first — and come back for the rest before Chapter 20.