Pairwise Preference Rubric (Response A vs Response B)

Evaluates two competitive model completions across strict alignment, instruction following, and conciseness.

Evaluate the two provided assistant completions (Response A and Response B) responding to the user prompt.

Evaluation Dimensions:
1. Instruction Adherence: Did the response follow all negative and positive constraints (word count limits, prohibited phrases, required formatting)?
2. Factual Grounding: Are all claims, figures, and technical assertions verifiable without hallucination?
3. Conciseness & Signal-to-Noise: Does the completion deliver direct value without redundant filler or conversational padding?

Output Schema:
- Preferred Response: [Response A / Response B / Tie]
- Justification: Cite explicit quoted evidence from both responses demonstrating why one output is strictly superior.

User Prompt & Responses:
[INSERT PROMPT AND RESPONSES HERE]