Framework

Boundary Value Analysis, and Why Governance Needs the Same Discipline

Boundary value analysis is a testing technique from software quality engineering, and it rests on a simple observation: defects cluster at the edges of valid ranges, not in the middle of them. If a field accepts values from 1 to 100, the interesting test cases are 0, 1, 100, and 101, not 50. The boundary is where things break.

The IIBA's Business Analysis Body of Knowledge extends a related idea to requirements themselves: atomic constraint satisfaction. A complex requirement is most reliably verified not as one composite judgment, but as a determination of whether each of its indivisible constituent constraints has been independently satisfied. Complexity hides at the seams between constraints, the same way defects hide at the boundary between valid and invalid values.

Most AI governance treats policy compliance as a single aggregate judgment, an overall risk score, a general sense that an action seems fine. That approach loses exactly the information boundary value analysis says matters most, which specific constraint, at which specific edge, was the one that mattered.

An architecture that evaluates governance, cost, and confidentiality as three independently satisfied constraints, rather than one blended score, is applying the same discipline to AI agent actions that boundary value analysis applies to test cases and atomic constraint satisfaction applies to requirements. Find the edge. Test the edge. Never average it away.