Constructs
What the pipeline tries to measure
| Construct | Definition | Core danger |
|---|---|---|
| Judgment retention | Whether the participant retains independent criteria and accountability after AI exposure. | AI substitutes for judgment. |
| AI overreliance | Uncritical adoption of AI claims without evidence, critique, or verification. | Fluent rubber-stamping. |
| Cognitive outsourcing | Delegation of diagnosis, criteria, tradeoff reasoning, or accountability to AI. | The person becomes a wrapper for AI output. |
| Critique quality | Ability to identify unsupported claims, missing assumptions, weak logic, and useful contributions. | Generic “AI can be wrong” caveats without substance. |
| Decision improvement | Whether the final decision improves in evidence, tradeoffs, implementation, and accountability. | Better prose without better reasoning. |
| Confidence calibration | Whether confidence changes track actual reasoning improvement and remaining uncertainty. | Confidence inflation after AI exposure. |