"The Five Camps of Data Modeling (and Which One You're Stuck In)" — Joe Reis

Why this is in the vault

Joe Reis names five historically isolated data-modeling traditions (Relational, Analytics, Application, ML/AI, Knowledge) as a free re-adaptation of Chapter 1 of his book Mixed Model Arts — the same source material the vault already tracked in full via the chapter-by-chapter backfill and the closing manifesto.

The core argument

Reis opens with a real failure story: an app team's JSON-blob product catalog broke both an analytics dashboard and an ML recommendation pipeline, because each team modeled the same data for its own consumer without looking across the architecture. He traces this to five separate professional traditions that grew up solving different problems and rarely talk to each other:

• Relational (Codd, 1970) — normal forms, referential integrity; blind spot is analytical queries and unstructured/vector data.
• Analytics (1990s, Inmon/Kimball/Data Vault) — schemas for OLTP-to-warehouse queries; blind spot is real-time streams, graphs, feature tables.
• Application (2000s-2010s NoSQL) — models around app access patterns (MongoDB, Kafka); blind spot is everyone downstream who inherits the resulting chaos.
• ML/AI — feature tables, embeddings; blind spot is point-in-time correctness and entity resolution, causing data leakage.
• Knowledge (library science, ontology, enterprise architecture) — asks what terms mean; historically peripheral, now pulled to center stage because LLM agents need to know what a "Customer" actually is.

His argument: three historical waves (operations-meets-analytics, big data/lakes, ML/AI) stacked rather than replaced each other, so a single transaction ("Order 808," a kettlebell purchase) now exists in six simultaneous representations across an architecture — and when the identity keys linking those representations break, the whole system falls apart. The prescription is cross-disciplinary literacy, not mastery of all five: keep a primary specialty, get competent in the rest.

Mapping against Ray Data Co

Medium. This reinforces, but doesn't extend, the strong mapping already filed for Reis's Aug 19 manifesto (2026-08-19-practical-data-modeling-mma-manifesto.md): the "unowned center" problem in phData's DIE hub-and-spoke map — the Governed Knowledge Graph / Fabric spoke nobody owns — is structurally the isolation problem this piece describes. This piece adds a naming layer worth borrowing directly: Murray's AWS/Neptune spoke reads as the Knowledge camp, Anderson's Snowflake spoke as the Analytics camp, Andrew's Anthropic spoke as the ML/AI camp — three of Reis's five camps mapping cleanly onto three named phData spokes that don't currently share vocabulary. That's a reusable naming convention for the founder's Fabric pitch ("we're not proposing a sixth camp, we're naming the seam between the five you already have"), not a new mechanical insight.

Honest caveat: this is a repackaging move, not new payload. The taxonomy substantially overlaps with two pieces already in the vault — the Feb 2026 MMA Ch1 chapter and the June 2026 "Turf Wars Are Over" manifesto — and the piece is explicitly framed as the first in a series of free/paid adaptations funneling toward the Mixed Model Arts ebook (annual-subscription gate, Amazon listing "coming very soon"). Treat it as a marketing-optimized restatement of a belief the vault already holds with more depth, not a fresh argument.

Related

• [[2026-02-18-practical-data-modeling-mma-ch1-full]] — the original MMA Chapter 1 this piece re-adapts for a wider (free-tier) audience
• [[2026-06-02-practical-data-modeling-turf-wars-cross-train]] — the closest thematic cousin: same camps-in-isolation critique, framed as MMA/cross-training instead of a five-camp taxonomy
• [[2026-08-19-practical-data-modeling-mma-manifesto]] — the strong-mapping note this piece reinforces (DIE Fabric / unowned semantic center)

Copyright note

Quotes ≤15 words, paraphrase otherwise. Source: Practical Data Modeling, Sep 28 2026 — view at https://practicaldatamodeling.substack.com/p/the-five-camps-of-data-modeling-and