{
  "predict": {
    "traces": [],
    "train": [],
    "demos": [],
    "signature": {
      "instructions": "Your objective is to carefully analyze highly localized, noisy, bilingual (Arabic/English) Saudi Arabian banking transaction logs to classify whether a given description represents a failed, reversed, refunded, rejected, or cancelled transaction. \n\nThese transaction logs are complex: they blend unstructured user memos, system-generated metadata, and legacy encoding artifacts. They frequently feature intra-sentence code-switching (mixing Arabic and English), inconsistent PII masking, and varied database serialization structures. Therefore, your analysis must go beyond simple keyword matching; it requires deep semantic understanding of financial jargon and the ability to disambiguate entity naming overlaps from actual transaction statuses.\n\nPay close attention to context and indicators of transaction failure, reversal, or refund in both languages. \nCommon Arabic indicators include: مرفوض، مرفضة، استرجاع، مرتجع، الغاء، ملغاة، رفض، استرداد، إعادة، فشل، عكس، تصحيح.\nCommon English indicators include: refund, reversed, reversal, failed, rejected, cancellation, correction.\n\nFor each transaction description, you must:\n1. Provide a clear, logical reasoning explaining why the text does or does not indicate a failed/reversed transaction, taking into account the semantic context.\n2. Determine the boolean classification (True if it is a failed, reversed, refunded, rejected, or cancelled transaction; False if it is a standard successful transaction).\n3. Assign a confidence level ('low', 'medium', or 'high') to your prediction based on the clarity and explicitness of the text.",
      "fields": [
        {
          "prefix": "Description:",
          "description": "The bank transaction description text"
        },
        {
          "prefix": "Reasoning:",
          "description": "Brief explanation of why this transaction is or is not failed"
        },
        {
          "prefix": "Is Failed Transaction:",
          "description": "True if the transaction is failed/reversed/refunded/rejected/cancelled"
        },
        {
          "prefix": "Confidence:",
          "description": "Confidence level of the classification"
        }
      ]
    },
    "lm": null
  },
  "metadata": {
    "dependency_versions": {
      "python": "3.12",
      "dspy": "3.1.3",
      "cloudpickle": "3.1"
    }
  }
}