Skip to main content
Use these commands before promoting governance changes.

​
1) Validate configuration

dazense validate
Checks consistency across:
  • dataset bundles
  • policy configuration
  • semantic model references

​
2) Run governance evaluation

dazense eval
Runs governance test cases from bundle eval_test_cases.

​
3) Run engine-based evaluation

dazense eval --engine
Runs Python-native policy checks (stronger than structural heuristics):
  • PII checks (including SELECT *)
  • table scope detection
  • time filter checks
  • business-rule matching

​
4) Run scorecard only

dazense eval --scorecard
Shows governance completeness checks (for example require_contract, bundle requirements, PII declarations).

​
5) Mode-based manual checks

For end-to-end UX validation, test each mode:
  1. none — no governance
  2. semantic — semantic model + business rules
  3. full — policy + contracts + bundle constraints
Use your mode switch script and start a new chat after each switch.

​
Minimum production bar

  • dazense validate passes
  • dazense eval passes
  • dazense eval --engine passes
  • PII prompts are blocked in full mode
  • Critical metrics (for example revenue) are consistent across supported LLMs