Name: stable, no temporary shorthand
Meaning: definition and example
Grain: what one row or value represents
Quality: nulls, enums, updates
Access: sensitivity classification
Consistent names do not create consistent semantics
Renaming “gmv_amt” to “transaction amount” improves readability but says nothing about tax, refunds, time, or grain. An AI-ready field needs a name, definition, example, type, grain, update time, and owner.
Naming should serve metrics and questions, not a tidy-looking vocabulary.
Remove hidden abbreviations and filters
A field named “status” means different things in order, payment, and logistics tables. “Active” can mean recent login, valid contract, or saleable product. Describe enums, default filters, and exceptions.
For amounts record currency, tax, discounts, and precision. For dates record time zone and business event.
Make metadata testable
Version the dictionary and test type, uniqueness, enum drift, null ratio, and freshness. BuildTable’s AI-friendly modeling direction can be evaluated for this work; exact metadata and publishing capabilities require project confirmation.
A field change should notify the owners of dependent metrics, reports, and agents.
Validate with real questions
Use ten business questions to check whether a business reader understands the description, the model selects the correct column, and the answer is explainable. Do not review naming in isolation from the question set.
Public references
Build an AI-ready data foundation
Download BuildTable or talk with us about your data modeling scenario.
Download BuildTable