Creating and Editing DQ Test Rules

New in Version 6.1

Profiling and creating rules requires the Rule Creator role or above.

Profiling a table shows you what its data actually looks like. From there, you can create a DQ test rule three ways: in plain English, by writing your own SQL, or by letting KDQ Agentic's AI assistant suggest a set of rules for you.


Profiling a table

Select an imported table and click Run Profile. KDQ Agentic scans each column and reports:

  • How complete it is — the share of empty values.

  • How varied it is — how many distinct values, and whether they're unique.

  • Its range — smallest and largest values, for numbers and dates.

  • Its shape — how well values match a common format (e.g. an email pattern).

  • Its common values — the top values and how often they appear.

Profiling samples large tables for efficiency, but always measures completeness exactly.


Creating a rule

Click New Rule on a profiled table to see the three ways to create a rule:

Plain English

  • Describe what you want, e.g. "the email column should never be more than 2% empty."

  • KDQ Agentic interprets it instantly and shows you the dimension, check, and threshold it understood, a confidence score, a suggested severity, and a short explanation of how it read your request. A low confidence score usually means the wording was ambiguous — try rephrasing.

  • Keep the conversation going with follow-ups like "also check the email is a valid format" or "change the threshold to 5%" — one description can turn into several related rules.

Manual SQL

  • Write your own check SQL directly.

  • KDQ Agentic annotates it with the dimension and logic it detects, and lets you validate the SQL against live data before saving.

AI-assisted generation

  • Choose Generate from Profile to have the AI assistant propose a set of rules for the table — the fastest way to get broad coverage.

  • Choose what the assistant sees: Full profile (the table's statistics — produces more rules, with thresholds grounded in your actual data) or Metadata only (just column names, types, and descriptions — fewer, more cautious rules, no statistics leave the workspace).

  • Review the summary before it's sent, and redact anything sensitive (e.g. sample values containing names).

  • Send it. Only your approved summary leaves your environment — never the raw records.

  • Review each suggested rule — confidence score, plain-language reasoning, and a suggested severity.

  • Approve the ones you want. Accepted rules become your draft rules; reject the rest.

⚠️ Only click Approve/submit once, and avoid using the browser back button to return to candidate review after submitting — doing either can create duplicate rules.

Every rule — however it's created — starts life as a draft. There's no separate approval step: accepting an AI suggestion simply makes it a draft rule you own.


Editing and managing draft rules

  • Open any draft rule to edit its logic, threshold, or SQL at any time.

  • Give each rule a severity (low / medium / high / critical) and a short business justification — a plain note on why the rule exists, so anyone reviewing it later understands the intent.

  • Mark the business priority of individual columns to tell the AI assistant which columns to cover first next time you generate rules.

  • Set a manual K Platform catalog reference on a table or column if the automatic link needs overriding.

  • Use bulk actions to delete, re-prioritise, or assign several rules to a schedule group at once.

🔒 Deleting a rule is restricted to Workspace Admins in the KDQ Agentic interface, even though Rule Creators can do everything else with a rule.


Change history

Version 6.1 ·

  • NEW   Profiling tables and creating DQ test rules in KDQ Agentic — plain English, manual SQL, and AI-assisted generation.

Last updated: