lang-quiz/prompt.txt

You analyze an Instagram-style Japanese language video.
The video is either Japanese-only or English+Japanese.

Goal: Create a compact learning JSON for a custom quiz website.
The website will ask the user questions; if the user is wrong or taps "Don't know",
we will show the explanation from this JSON and the original IG media.

TOKEN BUDGET:
Be concise. Do not duplicate explanations inside questions.

STRICT RULES:
1) Do not invent. Only include words/phrases/grammar that clearly appear in the video.
2) For every Japanese string containing kanji, provide a full hiragana reading in a separate field "kana".
3) Keep it small:
   - grammar: max 2
   - vocab: max 10
   - conversation lines: max 10
   - key_phrases: max 10
4) NO timestamps. NO evidence_quote. NO source fields. NO confidence fields.
5) Questions MUST reference items by id (do not repeat long explanations in questions).
6) Provide 1–3 questions per item (depending on usefulness). Prefer: cloze, multiple choice, match, register-choice.

OUTPUT:
Return ONLY valid JSON (no markdown). UTF-8 Japanese.

SCHEMA:

{
  "meta": {
    "mode": "ja_only|en+ja",
    "type": "grammar|vocab|conversation|mixed|unknown",
    "title_en": "short title (5-8 words max)"
  },
  "items": {
    "grammar": [
      {
        "id": "g1",
        "pattern": "string",
        "meaning_en": "one line",
        "use_note_en": "1-2 lines max",
        "register": "polite|neutral|casual|slang|formal|unknown",
        "example": {
          "jp": "string",
          "kana": "string (hiragana reading; required if jp has kanji)",
          "en": "string"
        }
      }
    ],
    "vocab": [
      {
        "id": "v1",
        "jp": "surface form",
        "kana": "hiragana reading (required if jp has kanji; for kana-only words keep as-is)",
        "meaning_en": "short",
        "register": "polite|neutral|casual|slang|formal|unknown",
        "note_en": "optional, 1 line max",
        "example": {
          "jp": "optional",
          "kana": "optional",
          "en": "optional"
        }
      }
    ],
    "conversation": [
      {
        "id": "c1",
        "jp": "exact line",
        "kana": "hiragana reading (required if jp has kanji)",
        "en": "translation",
        "register": "polite|neutral|casual|slang|mixed|unknown"
      }
    ],
    "key_phrases": [
      {
        "id": "k1",
        "jp": "phrase",
        "kana": "reading (hiragana if needed)",
        "meaning_en": "short",
        "when_to_use_en": "1-2 lines max",
        "register": "polite|neutral|casual|slang|formal|unknown"
      }
    ]
  },
  "quiz": [
    {
      "id": "q1",
      "targets": ["k1"], 
      "type": "mc_meaning|mc_register|cloze|match|choose_best_reply",
      "prompt_en": "string",
      "payload": {
        "sentence_jp": "optional",
        "sentence_kana": "optional",
        "blanked": "optional (use ____ for blank)",
        "options": ["A", "B", "C", "D"],
        "pairs": [{"left":"", "right":""}]
      },
      "answer": {
        "correct_index": 0,
        "correct_text": "optional (for non-mc types)"
      }
    }
  ],
  "ui_hints": {
    "recommended_order": ["g1","k1","k2","v1"],
    "show_first": "quiz",
    "explain_on_fail": true
  }
}

QUESTION GUIDELINES:
- For each grammar item: make at least 1 cloze question from the example, and 1 meaning or usage question.
- For vocab/key_phrases: make at least 1 meaning multiple-choice and optionally 1 register or “best situation” question.
- For conversation lines: optionally make “choose the best reply” or “what does this line mean” questions.
- Keep prompts short. Do not restate long explanations. The website will show explanations after.

Now analyze the video and output ONLY the JSON.