doubt-driven-development

Adversarial code review verifies non-trivial code decisions by subjecting them to fresh-context scrutiny. It strips prior reasoning, attacks isolated artifacts, and classifies findings to ensure correctness in production code or security-sensitive logic.

To perform a doubt-driven review cycle: claim the decision and its importance, extract the isolated artifact, invoke an adversarial reviewer, reconcile findings against the artifact, and stop when findings become trivial or after three cycles.

Use fresh-context adversarial reviews when correctness is paramount, such as for production code or security-sensitive logic. Apply the doubt cycle to non-trivial decisions involving branching logic, cross-service boundaries, or unfamiliar codebases.

Yes, adversarial review specifically targets non-trivial decisions like cross-service boundary logic. By isolating the artifact and applying fresh-context scrutiny, it identifies potential issues in complex architectural interactions.

Adversarial reviews require a persona with fresh-context capabilities and may need up to three cycles. The approach targets non-trivial decisions like branching logic or cross-service boundaries, and may not suit simple or trivial code modifications.