{
  "id": "extract_fields",
  "version": "1.0.0",
  "kind": "domain",
  "description": "Classify receipt OCR notes into structured receipt JSON. HTTP is unused when llm_role=classify.",
  "input_schema": {
    "type": "object",
    "properties": {}
  },
  "output_schema": {
    "type": "object",
    "description": "Receipt extraction from OCR notes and the goal. Always emit every required key. Use null when a value is not in the notes; do not invent. Set human_review_required when a receipt field is null or confidence is low.",
    "required": [
      "merchant",
      "amount",
      "currency",
      "date",
      "tax",
      "missing_information",
      "confidence",
      "human_review_required",
      "human_review_reason"
    ],
    "properties": {
      "merchant": {
        "type": ["string", "null"],
        "minLength": 1,
        "maxLength": 120,
        "description": "Merchant name as printed, or null if missing. Copy from OCR notes; do not invent."
      },
      "amount": {
        "type": ["number", "null"],
        "minimum": 0,
        "description": "Total charged, or null if missing. Use the receipt total, not tax alone."
      },
      "currency": {
        "type": ["string", "null"],
        "enum": ["USD", "AUD", "EUR", "GBP"],
        "description": "ISO 4217 code when stated, or null if the receipt does not name a currency."
      },
      "date": {
        "type": ["string", "null"],
        "pattern": "^[0-9]{4}-[0-9]{2}-[0-9]{2}$",
        "description": "Purchase date as YYYY-MM-DD, or null if missing. Convert printed dates to this form."
      },
      "tax": {
        "type": ["number", "null"],
        "minimum": 0,
        "description": "Tax amount when printed, or null if not stated. Do not invent."
      },
      "missing_information": {
        "type": "array",
        "items": { "type": "string" },
        "description": "Property names that are null or too unclear to extract. Empty array if complete."
      },
      "confidence": {
        "type": "number",
        "minimum": 0,
        "maximum": 1,
        "description": "Confidence score from 0 to 1 for the extraction as a whole."
      },
      "human_review_required": {
        "type": "boolean",
        "description": "True when a receipt field is null, confidence is low, or the notes are ambiguous."
      },
      "human_review_reason": {
        "type": ["string", "null"],
        "description": "Why review is required, or null when human_review_required is false."
      }
    }
  },
  "invoke": {
    "method": "POST",
    "url": "https://api.internal/receipts/extract",
    "auth": "domain-oauth"
  },
  "snippet": null,
  "owner": "ops",
  "status": "published"
}