Model Reviews and Modelling Standards

Model Reviews and Modelling Standards

Naming conventions, review cadence, model registry — the boring discipline that compounds.

Why it matters

Every model is read 10× more than written. Naming + standards are the lowest-cost, highest-leverage discipline DMBOK names. Skip them and every new joiner pays the comprehension tax.

Going deeper

The smallest standards pack that actually moves the needle:

• Naming — snake_case, singular nouns, no abbreviations except an enterprise abbreviation list (≤ 30 entries).
• Keys — surrogate id BIGINT PK on every entity; natural keys in a UNIQUE constraint, not the PK.
• Timestamps — created_at, updated_at, deleted_at (soft-delete), UTC.
• Review — every model change reviewed by a steward + reasoner / linter; approved before merge.
• Registry — a single searchable place where the current model lives (dbt docs, Atlan, DataHub).

A team that ships these five wins back 20–30 % of analyst onboarding time. There is no cheaper governance investment.

Modelling standards are the side of the road you drive on.

It doesn't matter much which side — left or right — but it matters enormously that the whole country picks one. The cost of inconsistency is paid in head-on collisions; the cost of consistency is paid once, by the first driver to learn the rule.

Same with created_at vs dt_created vs creation_date: pick one and enforce it. The cost is one design meeting; the savings are every future analyst who can guess the column name on the first try.