self-eval

To get an honest quality score for completed AI work, use a calibrated evaluation that separates task ambition from execution quality and applies a devil's advocate review to counter score inflation. This prevents vague single-number judgments by forcing skeptical analysis.

A two-axis rubric for AI task evaluation scores both the difficulty of what was attempted and how well the execution was performed. Use this calibrated scoring during post-task reflections, code review wrap-ups, and debugging session retrospectives for an unbiased assessment.

To run a devil's advocate review during AI session retrospectives, the evaluation process forces a skeptical pass that argues for both lower and higher scores before finalizing the result. This mandatory anti-inflation checking ensures the final calibrated score remains unbiased.

Yes, you can track AI assessment quality scores across multiple sessions through score persistence. The system records calibrated scores over time and flags repeated clustering or trends, which helps identify potential bias in your retrospective evaluations.

No, this self-evaluation approach requires no external dependencies or components to function. You simply ask the assistant to evaluate your completed work and return a quality assessment using the built-in ambition-and-execution scoring rubric.

Your AI code review assessment needs to separate ambition from execution because a single-number judgment often obscures whether a high score reflects an easy task done well or a complex task done poorly. A two-axis rubric provides a calibrated score reflecting both.