Provenance records where a result or data movement came from and which steps produced it. It turns an unexplained answer into an inspectable explanation: why an account reached a sales organization, why traffic passed a firewall, or where a test fixture originated. section 8.6 section 8.2
Two practical designs make provenance visible:
- Return a result that carries both the decision and its explanation, such as matched, expected, and found values. section 8.6
- Generate integration-test inputs inside the test, so assertions connect to declared data rather than an unexplained hard-coded file. section 12.2
- Preserve the same rule data while changing the interpreter, allowing increasingly detailed explanations without changing the rules themselves. section 8.6
Explaining a failed rule
A rule may require country=US, while an account contains country=BE. Instead of returning only false, the evaluator returns matched=false, expected=(country=US), and found=(country=BE). Compound AND, OR, and NOT rules can preserve the same structure across the whole decision.
Making test data traceable
Rather than loading path/to/report.csv, a test generates rows, selects the relevant subset, serializes the rows, uploads the resulting file, and compares ingested accounts with expected identifiers. The same approach can scale to much larger inputs.

















