Skip to main content

Data dictionary

The data dictionary records what your data means: descriptions for every dataset and column, plus a glossary of defined metrics. It is also a drift detector -- when someone builds a card whose name matches a defined metric but whose SQL differs, the editor warns them before two versions of the same number reach your viewers.

Datasets and columns​

Open Data dictionary in the library sidebar. Every dataset appears automatically with the columns from its last run, and a coverage figure ("1 of 2 columns described") shows where documentation is missing. Anyone can read the dictionary; describing things requires the Builder tier (T4) or above.

Descriptions do double duty: in the card builder, a described column shows its definition beneath the column picker once selected, so builders see what a field means exactly where they use it.

Sensitive datasets follow the same visibility rules as their data: if you cannot open the dataset, you will not see its dictionary entries either.

The metrics glossary​

A metric entry is a name, a plain-words definition, and optionally the canonical SQL that computes it. Entries come from two places:

  • Manual definitions -- a Builder adds a term like "Active customer" with its agreed meaning.
  • Certified work -- when a direct-SQL card (or a dashboard containing them) is certified, its SQL is registered automatically as the canonical definition under the card's name. Certification is the gate: uncertified experiments never enter the dictionary.

Conflict warnings​

Whenever a card is saved -- and live in the editor as you type -- Flynt checks the card's name and its SQL output column names against the glossary. If a name matches a defined metric:

  • Definitions agree: a small confirmation that your card matches the certified definition. Formatting, capitalisation and alias-order differences do not count as disagreement.
  • Definitions differ: a warning that a certified metric with this name already exists, with a link to the dictionary. You can always still save -- the dictionary informs, it never blocks.

Good to know​

  • The check matches on names. A card that computes the same business metric under a completely different name is not detected -- naming discipline is what powers the warnings.
  • Editing a harvested entry's SQL by hand marks it as manual; the next certification of the source card reclaims the definition.
  • Deleting a metric removes its definition and its warnings; certifying the source card again restores it.