An architect is designing prompts for a Claude-powered code review assistant. Initial testing reveals that Claude sometimes generates insightful analysis and sometimes produces shallow bullet-point summaries of the same code. The temperature is set to 0.7. What prompt engineering approach most reliably improves response depth consistency?

D is correct because response depth inconsistency is a structural problem best solved with structural constraints. Defining required sections with depth criteria creates a template that Claude fills consistently, preventing the collapse into shallow bullet points. Including a negative example explicitly teaches Claude what 'not to do,' which is particularly effective for preventing known failure modes. A is wrong because shallow responses at temperature 0.7 are not primarily a sampling temperature problem — they are a structural guidance problem. Temperature reduction would make outputs more consistent but would not inherently make them deeper without structural scaffolding. C is wrong because more domain knowledge in the system prompt does not guarantee depth — it gives Claude more material but does not constrain how that material is applied in responses. B is wrong because selecting the longest response conflates verbosity with depth and creates unnecessary API cost; a structurally guided single call is more reliable than five unguided calls.