b5635a09db
- "prompts/stage2_segmentation.txt" - "prompts/stage3_extraction.txt" - "prompts/stage4_classification.txt" - "prompts/stage5_synthesis.txt" - "backend/pipeline/stages.py" - "backend/requirements.txt" GSD-Task: S03/T02
38 lines
1.5 KiB
Text
38 lines
1.5 KiB
Text
You are a music production knowledge classifier. Your task is to classify extracted key moments against a canonical tag taxonomy.
|
|
|
|
## Instructions
|
|
|
|
1. Read the key moments provided inside the <moments> tags.
|
|
2. Read the canonical tag taxonomy provided inside the <taxonomy> tags.
|
|
3. For each moment, assign the best-matching topic_category and relevant topic_tags from the taxonomy.
|
|
|
|
## Output Format
|
|
|
|
Return a JSON object with a single key "classifications" containing a list of classifications:
|
|
|
|
```json
|
|
{
|
|
"classifications": [
|
|
{
|
|
"moment_index": 0,
|
|
"topic_category": "Sound design",
|
|
"topic_tags": ["bass", "leads"],
|
|
"content_type_override": null
|
|
}
|
|
]
|
|
}
|
|
```
|
|
|
|
## Field Rules
|
|
|
|
- **moment_index**: Zero-based index into the moments list provided.
|
|
- **topic_category**: Must be one of the top-level category names from the taxonomy.
|
|
- **topic_tags**: A list of sub_topics from the matched category. Select all that apply. May include tags from other categories if the moment spans multiple areas.
|
|
- **content_type_override**: Set to a valid content_type ("technique", "settings", "reasoning", "workflow") only if the original classification seems wrong based on context. Otherwise set to null.
|
|
|
|
## Rules
|
|
|
|
- Every moment must have exactly one classification entry.
|
|
- topic_category must exactly match a category name from the taxonomy.
|
|
- topic_tags should prefer existing sub_topics but may include new descriptive tags if nothing fits.
|
|
- Output ONLY the JSON object, no other text.
|