Response Format

Success Response

{
  "prediction": {
    "id": "6a2a7242-f22e-4629-9c6a-895618ae26e8",
    "status": {
      "code": 0,
      "message": "ok"
    },
    "metadata": {
      "num_columns": 3,
      "num_predictions": 1,
      "num_query_rows": 1,
      "num_rows": 3
    },
    "predictions": [
      {
        "PRODUCT": "Wireless Headset",
        "SALESGROUP": [
          {
            "prediction": "Enterprise",
            "confidence": 0.87,
            "confidence_interval": null
          }
        ]
      }
    ],
    "explanations": {
      "top_column_scores": [
        {
          "CATEGORY": 0.74,
          "PRODUCT": 0.26
        }
      ],
      "top_relevant_context_rows": [
        [0, 1]
      ]
    }
  },
  "delay": 312.5,
  "samplingMetadata": {
    "downsampled": false,
    "sampledFrom": 2,
    "sampledTo": 2,
    "strategy": "random"
  },
  "aiApiRequestPayload": { "...": "full payload forwarded to the model" },
  "aiApiResponsePayload": { "...": "raw model response" }
}

Response Fields

• prediction: The prediction result object
  • id: Unique prediction identifier
  • status: Model status (code: 0 = success)
  • metadata: Summary of the prediction run
    • num_columns: Number of feature columns
    • num_predictions: Number of predicted values returned
    • num_query_rows: Number of rows that were predicted
    • num_rows: Total rows sent to the model (context + query)
  • predictions: Array of results, one per query row. Each element contains the predicted column(s) with prediction, confidence, and confidence_interval. When index_column is set, the index value is also included.
  • explanations: Column importance scores and relevant context row indices. See Explainability.
• delay: End-to-end processing time in milliseconds
• samplingMetadata: Context downsampling details (always present)
  • downsampled (boolean): true when context was reduced to fit the model limit
  • sampledFrom (integer): Context row count received from the client
  • sampledTo (integer): Context row count actually sent to the model (capped at 2,048)
  • strategy (string): Sampling strategy used. Currently always "random"
• aiApiRequestPayload: The exact payload forwarded to the RPT model. Useful for debugging.
• aiApiResponsePayload: The raw response received from the RPT model.

Multi-Column Prediction

To predict more than one column, set [PREDICT] on each target column in the query row:

{
  "rows": [
    {
      "PRODUCT": "Laptop Pro",
      "QUANTITY": 45,
      "REVENUE": 44999.55
    },
    {
      "PRODUCT": "Standing Desk",
      "QUANTITY": 28,
      "REVENUE": 15372.00
    },
    {
      "PRODUCT": "Wireless Headset",
      "QUANTITY": "[PREDICT]",
      "REVENUE": "[PREDICT]"
    }
  ]
}

For details on context row downsampling behavior, see Limits & Quotas.