Using a separate role or model instance to evaluate an implementer's work.
Implementer–reviewer separation reduces self-confirmation by giving evaluation to a worker with different context, incentives, or instructions. The reviewer checks against explicit requirements and evidence rather than trusting the implementer's summary.