# Business feature experiment

- User problem and falsifiable hypothesis:
- Control / treatment; which behavior changes:
- Eligible population and randomization unit:
- Exclusions, risk limits and protected tenants:
- Stable identity and assignment persistence:
- Actual exposure event / timestamp / feature version:
- Primary outcome, denominator, deduplication and conversion window:
- Guardrails, threshold, owner and containment action:
- Baseline, minimum worthwhile effect and sample/duration rationale:
- Planned analysis and uncertainty method:
- Data-quality checks: missing exposure, cross-arm identities, duplicate conversions, missing events, unequal allocation:
- Continue / expand / stop / inconclusive decision criteria:
- Final decision, owner, evidence and cleanup:

Evaluation/assignment is not proof that the user saw a feature. Verify successful delivery and the actual analytics pipeline. Synthetic fixture output demonstrates validation, not customer impact or a winning variant.