Are the Levels of Data Modeling Outdated? (Redux)

Expanded revisit of whether the CDM/LDM/PDM hierarchy still applies. Reis argues the levels are not outdated but should be treated as a toolkit, not a mandatory ceremony. Developers already practice them implicitly — just not always intentionally or collaboratively.

Key positions:

• AI agents accelerate but don’t replace modeling judgment — they excel at mechanical translation between levels but cannot validate business correctness
• Polyglot persistence (NoSQL, vectors, event streams) has blurred the logical/physical boundary, making conceptual alignment more critical than ever
• Data contracts use the CDM as their foundation — without conceptual agreement, distributed data products lack interoperability
• When to skip levels: well-understood narrow domains can skip formal CDM; brownfield work should start at PDM and reverse-engineer upward; exploratory work can ignore all levels initially

Anti-patterns cataloged: “throw it all in one DB,” the “what’s a customer” game, and the “perfect schema” trap (over-normalized model requiring 17 joins).

RDCO relevance

Practical guidance for client engagements. The “levels as toolkit” framing helps us calibrate formality to client maturity. The data contracts angle connects to how we structure dbt project handoffs between producer/consumer teams.