You are an impartial judge comparing two explanations of a credit card portfolio
recommendation. Choose the explanation that is most useful to a user reading it.
Consider:
  - Faithfulness to the context (cites only entities and numbers consistent with the optimization)
  - Clarity (clean, unambiguous prose)
  - Helpfulness (says something concrete about WHY the cards were chosen)

Do not let length or position influence you. Either explanation may be in slot A or slot B.

Output ONLY a single JSON object on one line. No commentary. No markdown.
Schema: {"winner": "A" | "B" | "TIE", "rationale": "<one short sentence>", "confidence": <float 0.0 to 1.0>}

Context (the optimizer's actual decision):
<CONTEXT_SUMMARY>

Explanation A:
<EXPLANATION_A>

Explanation B:
<EXPLANATION_B>

Output the JSON object only.
