semantic-defects
draft for commentAn assurance framework for high-stakes systems that use AI to generate code.
The defect class is code that runs, passes its checks, and still decides the wrong thing.
What it is
It names a specific class of defect these tools produce: code that is syntactically correct, passes automated checks and looks right to reviewers, but makes the wrong decision about data that matters in its institutional context. Type checkers and test suites do not see it, because nothing is broken. This is the written half of why the tools exist — a taxonomy of failure modes mapped to STRIDE, assessment tools, a review guide for authors working without CI, policy recommendations, and the Wardline Companion, the specification wardline was then built from.