You are given an existing list of agent evaluation questions used to measure adherence to provided knowledge bases (KBs). You now have a new evaluation question and three possible integration outputs (scenarios):

1. IGNORE:
    - Appropriate ONLY if the new question almost fully matches exactly or precisely represents a subset covered by an existing question.

2. UPDATE_EXISTING:
    - Appropriate ONLY if the new question overlaps strongly with an existing one, allowing the two to be cleanly merged into a concise, atomic question maintaining Yes/No binary clarity WITHOUT conjunctions ('and', 'or').

3. CREATE_NEW:
    - Appropriate ONLY if the new question introduces a distinctly new scenario or performance evaluation aspect not sufficiently covered by existing ones.

Your task is evaluating provided scenario outputs (IGNORE, UPDATE_EXISTING, and CREATE_NEW) and carefully selecting the optimal scenario according to criteria:

- Atomicity: Each question must center strictly on a single evaluative aspect.
- Functionality: Clear Yes/No logic based on concrete KB instructions and agents' actions.
- Simplicity: Questions should remain straightforward and not overly complex.

Existing questions:
```
{current_questions}
```

New Evaluation Question:
```
{new_question}
```

Output from IGNORE scenario:
```
{ignore_output}
```

Output from UPDATE_EXISTING scenario:
```
{update_output}
```

Output from CREATE_NEW scenario:
```
{create_new_output}
```

Output format (JSON):
{{
    "reasoning": "<Rigorous, intelligent reasoning about why your chosen scenario best meets the criteria of atomicity, functionality, and simplicity.>",
    "decision": "<IGNORE / UPDATE_EXISTING / CREATE_NEW>"
}}

Example:
```
New question: "When customers were unsure about their subscription details, did the agent accurately clarify the customer's active subscriptions?"

IGNORE output says a similar and less specific question exists like - "if the agent was unsure about the any of their profile details, did the agent accurately clarify those details".
UPDATE_EXISTING merges subscription-related questions distinctly.
CREATE_NEW introduces a completely separate question.

A potential output:
{{
    "reasoning": "The IGNORE scenario here excellently covers the new question because the subscription is also a part of the profile detail of the customer. It keeps atomicity by clearly focusing on profile details clarification action without conjunction. UPDATE_EXISTING violates simplicity and saturation principles, CREATE_NEW unnecessarily duplicates a related question.",
    "decision": "IGNORE"
}}
```

Only output the JSON described above and nothing else: