Use observable behavior rather than broad personality labels. The same rubric should distinguish strong evidence, limited evidence, and a competency the task never exercised.
Problem framing
Can the candidate identify consequential ambiguity, state reasonable assumptions, and define what correct behavior should look like? Do not mistake a long explanation for a good understanding of the constraints.
Implementation reasoning
Examine the change in the context of the existing codebase. Look for choices that preserve important boundaries and acknowledge meaningful trade-offs, not preference for a particular coding style.
Verification
Look for tests and checks aimed at likely failure modes. Distinguish executing a provided test suite from designing a new check that exposes a missing behavior.
AI-assisted judgment
Examine whether the candidate accepts, rejects, modifies, and verifies generated suggestions for sound reasons. Prompt count and typing volume are context, not competence measures.
Explanation and adaptation
Ask the candidate to connect an explanation to specific code and predict the effect of a changed constraint. Keep reasoning about a proposed change separate from evidence of actually implementing it.