Batch Extraction Schema for Repeatable Work

{
  "model": "gpt-6-luna",
  "reasoning": {"effort": "low"},
  "input": [
    {
      "role": "developer",
      "content": "Extract fields from {{document-type}} documents. Copy values exactly as written, do not normalise units or dates, and return null for any field not present rather than guessing. If the document is not a {{document-type}}, set document_ok to false and leave the other fields null."
    },
    {
      "role": "user",
      "content": "{{document-text}}"
    }
  ],
  "text": {
    "format": {
      "type": "json_schema",
      "name": "extraction",
      "strict": true,
      "schema": {
        "type": "object",
        "properties": {
          "document_ok": {"type": "boolean"},
          {{field-schema}}
        },
        "required": ["document_ok", {{required-fields}}],
        "additionalProperties": false
      }
    }
  }
}

Run on the Responses API so the schema is enforced. Keep effort at low for extraction; raise it only for documents where a field must be inferred rather than copied. Normalise dates, currencies and units in code after extraction, where the rules are yours and testable.