A calibrated judge has already decided each of the questions below about the test case. Your job is to explain WHAT IN THE TEST CASE made each outcome come out the way it did. Return a JSON object with exactly one field, `"reason"`.

Rules:
1. Cover EVERY question, in the order given, with one sentence or clause each. Do not skip any. For a question marked "not applicable", say what about the test case made it not apply.
2. Ground every point in the test case: quote or closely paraphrase the specific span of the fields in `Test Case` that made the outcome hold or fail. "The response contradicts the expected output" is NOT acceptable; "The response says the answer 'depends' and offers the dog, while the expected output is only 'The cat'" is.
3. Never justify anything by a number or a level of certainty. Do not mention scores, probabilities, percentages, "confidence", "the judge was sure", or "high/low". Where an outcome is "unclear" or "leaning", explain WHAT the evidence is ambiguous about, not how ambiguous it is.
4. Do not open or close with an overall verdict such as "overall the output is good/bad". The score already says that.
5. Do not restate the outcome word ("clearly holds", "likely fails", ...). Turn it into evidence.
{% if multimodal %}{{ _fragments.multimodal_input_rules }}{% endif %}

Bad reason (parrots outcomes, no evidence):
"The first check clearly fails and the second is unclear. The response was judged mostly wrong with good confidence, and it was partially answered."

Good reason (every question, every point cites the test case):
"The response entertains the dog as an answer while the expected output is only 'The cat', so it introduces a conflicting claim. It does name the cat, but hedges with 'it depends' rather than committing, so the expected detail is only partly delivered. Against the expected answer it lands closer to mostly wrong than mostly right for that same reason. The user asked a direct question and the response engaged with it but did not settle it, so it was partially answered rather than declined."

Only return valid JSON. Do **not** include any extra commentary or text.

---

Questions and outcomes (in order):
{% for item in outcomes %}
{{ loop.index }}. [{{ item.type }}] {{ item.question }}
{% if item.levels %}   Levels (worst to best): {{ item.levels }}
{% endif %}{% if item.options %}   Options: {{ item.options }}
{% endif %}   Outcome: {{ item.outcome }}
{% endfor %}
Test Case:
{{ test_case_content }}

Parameters:
{{ parameters }}

---
**Example JSON:**
{
  "reason": "your grounded, question-by-question explanation here"
}

JSON:
