"Week 1 — Choosing a Data Modeling Approach That Fits the Problem" — @practicaldatamodeling
Why this is in the vault
Launch post for a live re-read/discussion cohort of Mixed Model Arts (Chapters 1–3, discussion target Oct 21) — not a new chapter, but the first genuinely new argument since the original chapter backfill: it names the specific failure mode where an operational "closed" state and an analytical "resolved" state silently diverge, and frames the fix as a question of what representations must agree on, not which modeling camp wins.
The core argument
Reis opens with a familiar failure: a support app that works fine for its own users still produces a mess downstream, because different consumers (the agent closing tickets, the analyst measuring resolution time, a recommendation system clustering complaints) need different things from the same underlying business reality. Chapters 1–3 introduce the modeling "camps" and warn against treating a preferred approach as the answer before the question is understood.
The running example — borrowed forward from Chapter 12 — is an online retailer's support operation spanning email, chat, and phone. The concrete tension he poses: a closed ticket is a valid application state, but it doesn't necessarily answer the business question of whether the customer's problem actually went away. The discussion questions push on this directly: which modeling camp is your first instinct and what would it miss here; what must the application and analytics representations agree on for their answers to stay coherent; which assumption about "closed" would you actually go test with an agent or customer before building a report on top of it.
The exercise (sketch the app plus a resolution-time report, write the one question each serves, circle the one definition they must share) is a compact version of Reis's recurring thesis across the book: data debt isn't created by the tool choice, it's created by unexamined definitional choices made before anyone calls it "data modeling."
Mapping against Ray Data Co
This is the exact shape of the semantic-layer disagreement RDCO/phData engagements run into: an operational system's state field (ticket status, deal stage, claim status) gets reused as the analytics answer to a business question it was never designed to answer, and the gap surfaces downstream as a trust problem in the report rather than a definition problem in the source. Reis's discussion-question framing — "what must the two representations agree on" — is a usable diagnostic to bring into a data-modeling engagement kickoff rather than jumping straight to schema design.
Related
- [[2026-02-23-practical-data-modeling-mma-ch2-what-is-data-modeling]]
- [[2026-02-25-practical-data-modeling-mma-ch3-no-free-lunch]]
- [[2026-04-09-practical-data-modeling-mma-ch12-synthesis]]
- [[2026-08-19-practical-data-modeling-mma-manifesto]]