Build an analytics workflow for CSV and SQL with verified calculations, provenance, privacy controls, and a reproducible executive report.
Scenario
An analyst receives a CSV file and a SQL extract and must prepare findings for leadership. AI helps explore the data, but every number and claim must be verified and reproducible.
Outcome: Data quality, field semantics, and privacy risks are known before analysis begins.
Tasks
Checks
Outcome: Hypotheses and required calculations are separated from AI-generated narrative.
Tasks
Checks
Outcome: Results are reproducible independently of the model.
Tasks
Checks
Outcome: The narrative stays within the boundaries of the evidence.
Tasks
Checks
Assessment rubric
Schema, definitions, privacy, and quality rules are explicit.
Insufficient
Fields and metrics are ambiguous.
Competent
Definitions and checks are documented.
Strong
A reusable metric layer and automated data tests are implemented.
Evidence required
Key figures are reproducible with deterministic code.
Insufficient
The model generates figures in prose.
Competent
Key metrics are verified.
Strong
Independent reconciliation and regression checks are in place.
Evidence required
Claims are linked to a query, formula, or source slice.
Insufficient
Conclusions lack evidence.
Competent
Every key claim has provenance.
Strong
Lineage is generated automatically.
Evidence required
The report includes caveats and confidence and stays within the evidence.
Insufficient
Unsupported claims are present.
Competent
Claims, caveats, and uncertainty are explicit.
Strong
Adversarial review and an audience-specific summary are included.
Evidence required