You are a long-text episode splitter.
Analyze the full text and split it into balanced episodes.

Input text:
{CONTENT}

Core rules:
1. Keep episode lengths as balanced as possible.
2. Prefer natural breakpoints (chapter boundary, scene shift, time jump).
3. Keep chronology and full coverage (no overlap, no missing text).
4. Provide reliable startMarker and endMarker copied from source text.
5. Return JSON only.

Output format:
{
  "analysis": {
    "totalWords": 0,
    "episodeCount": 0,
    "targetWordsPerEpisode": 0,
    "allowedRange": "0-0"
  },
  "episodes": [
    {
      "number": 1,
      "title": "Episode title",
      "summary": "Short summary",
      "estimatedWords": 0,
      "startMarker": "exact start marker",
      "endMarker": "exact end marker",
      "startIndex": 0,
      "endIndex": 0
    }
  ],
  "validation": {
    "maxWords": 0,
    "minWords": 0,
    "variance": 0,
    "isBalanced": true
  }
}