Datasource dictionary
Scope
- Measures and calculations: concise default for metric-focused requests.
- Selected fields: best when the user names a business process or subset.
- All accessible fields: group by table/category and report complete coverage or explicit truncation.
Per-field record
Include applicable properties:
| Property | Guidance |
|---|---|
| Display name | returned Tableau caption/name |
| Business definition | confirmed description or confirm with data owner |
| Technical meaning | type, role, aggregation, table, and formula classification |
| Grain/unit | only when known |
| Directionality | favorable direction only with business evidence |
| Inputs | referenced fields for a calculation |
| Usage | visible/metadata-supported role, not guessed popularity |
| Caveats | null, filter, LOD, table-calc, currency, denominator, or privacy issue |
| Evidence status | observed, inferred, or unknown |
Never define a field by restating its name. Explain formulas in plain language without losing meaningful conditions, aggregation, fixed grain, addressing/partitioning, or null/zero behavior. Do not claim a table calculation's result without knowing the view context.
Sample/value enrichment
Only after explicit authorization, use a bounded query to obtain the specific range, example categories, date span, or magnitude needed. Record filters, aggregation, row limit, and timestamp. Avoid sensitive/free-text/identifier samples. A sample is not a full distribution and does not establish typicality without an appropriate method.
Large schemas
Group related variants as metric families. Give fuller definitions to decision-critical measures and calculations, and concise records to secondary fields. Do not invent priorities; use visible usage, user intent, descriptions, or calculation dependencies as evidence.