doubt-driven-development

Adversarial code review is a formal process to expose hidden assumptions and counterarguments in high-stakes development. It enacts an adversarial reviewer mindset that actively seeks contradictions and edge cases to disprove assumptions before they become firm conclusions.

To run an adversarial review, apply the five-step CLAIM, EXTRACT, DOUBT, RECONCILE, and STOP workflow. This structured doubt cycle forces explicit reasoning and uses adversarial prompts to stress-test outputs and surface flaws early.

Use an adversarial review process when correctness outweighs speed in high-stakes development or architecture choices. It is specifically designed for non-trivial decisions spanning modules, teams, or systems where confirmation bias poses significant risk.

Yes, doubt-driven development supports cross-context governance for decisions spanning modules, teams, or systems. It provides explicit artifacts, contracts, and outcomes to ensure structured reasoning across complex architectural boundaries.

Yes, the adversarial review workflow includes optional cross-model review capabilities for risk analysis. This allows you to leverage multiple perspectives to further stress-test non-trivial decisions and validate assumptions.

The formal doubt cycle prioritizes correctness over speed, making it less suitable for rapid, low-stakes development. Its structured five-step workflow requires explicit reasoning and artifact generation, which introduces overhead not warranted for trivial tasks.