Task types

Supported type values and how to shape label_spec and interpret consensus results.

Type	label_spec	Use case
image_classification	question + options[]	Categorize images at data_url
text_classification	question + options[]	Categorize text fetched from data_url
sentiment	question + options (e.g. positive / negative / neutral)	Sentiment
moderation	question + options (e.g. safe / unsafe + sub-categories)	Content safety

image_classification

{
  "type": "image_classification",
  "data_url": "https://cdn.example.com/photo.jpg",
  "label_spec": {
    "question": "Which product line?",
    "options": ["electronics", "apparel", "food"]
  }
}

Consensus result (illustrative):

{
  "consensus_label": "electronics",
  "confidence": 0.6666666666666666,
  "worker_count": 3,
  "agreement_count": 2
}

text_classification

{
  "label_spec": {
    "question": "Ticket priority?",
    "options": ["P0", "P1", "P2", "P3"]
  }
}

sentiment

{
  "label_spec": {
    "question": "Sentiment?",
    "options": ["positive", "negative", "neutral"]
  }
}

moderation

{
  "label_spec": {
    "question": "Is this content safe for the brand?",
    "options": ["safe", "unsafe", "review"]
  }
}

tip

Always provide at least two options. The consensus engine compares worker answers as strings; keep option labels stable across tasks for best agreement metrics.