🧠 Quiz Batch

German WWTBAM question bank transformed into deterministic Eleven3 scripts and rendered with the selected voice.

📦 Outputs

• questions: 300.
• tts_items: 600.
• prompts: 300.
• answers: 300.
• voice: nsiEp1alHkdiizNqCK2X.
• model: eleven_v3.
• audio: batches/audio/.
• manifest: batches/wwtbam-de-eleven3-audio-manifest.json.

🧭 Pages

• 📦 Batch Files: artifact map, schema, naming, shuffle, and validation shape.
• 🎛️ Prompt Stack: ear-first rewrite rules and canonical TTS templates.
• 🏷️ Tag Library: safe Eleven v3 tag groups and combo formulas.
• 🗣️ Voice Batches: generation route, concurrency policy, and audio cache contract.

🎭 Voice Profile

• prompt_open: [announcer, suspenseful].
• options: [clearly, curious].
• decision_gate: [pause] then [inviting].
• reveal: [confident, warm].
• bridge: [pleased, encouraging].

🛡️ Invariants

• first YAML answer remains the canonical correct answer.
• answer order is deterministically shuffled per category/question.
• correct answer never lands on A.
• prompt lines never reveal the answer.
• generated text contains no raw SSML, markdown emphasis, placeholders, or raw temperature symbols.
• MP3 cache validates by count, byte size, and ID3/frame-sync magic.