Optional Settings

ZeroTrain uses an Options object to control how models are executed and how results are returned. All settings are optional, and the system provides sensible defaults when they are not specified. These settings allow you to adapt behavior without modifying the underlying model or input data.

Options Structure

{
  "Options": {
    "DataSettings": { ... },
    "ContextOptions": { ... },
    "ConfidencePolicy": { ... },
    "OutputSettings": { ... }
  }
}

Data Settings

DataSettings define how the model is interpreted from structured data.

{
  "ActionColumnName": "Action",
  "PriorityColumnName": "Weight",
  "WeightColumnName": "Weight",
  "DecisionIdColumnName": "ID",
  "RowFilter": "IsEnabled = true",
  "ModelName": "CallCenterRoutingModel"
}

• ActionColumnName – Column that defines the resulting action
• WeightColumnName – Column used for scoring and prioritization
• DecisionIdColumnName – Unique identifier for each rule
• RowFilter – Filters which rows are considered during evaluation
• ModelName – Logical name for the model

Context Options

ContextOptions allow you to pass metadata through the execution pipeline.

{
  "CustomTag": "Call-Session-2026-0045"
}

This value is returned in the response header and can be used for tracking, correlation, or logging.

Confidence Policy

ConfidencePolicy controls how alignment strength is calculated across evaluated rows.

{
  "FitWeight": 1,
  "SeparationWeight": 0.03
}

• FitWeight – Emphasizes how well a row matches the input
• SeparationWeight – Adjusts how distinct a result is from other candidates

These values influence ranking behavior but do not change the deterministic nature of the result.

Output Settings

OutputSettings control how results and trace data are returned.

{
  "ExecutionTraceMode": "Detailed",
  "IncludeMatchedRow": true,
  "MatchedRowColumns": ["Weight", "HasPaymentFailure"]
}

• ExecutionTraceMode – Controls trace format (Compact or Detailed)
• IncludeMatchedRow – Includes selected row data in the result
• MatchedRowColumns – Specifies which columns are returned

Row Filtering

RowFilter allows dynamic inclusion or exclusion of model rows at runtime.

This enables scenarios such as disabling rules, segmenting logic, or controlling active decision sets without modifying the model structure.

Consistency

Settings do not introduce randomness. For the same input, model, and configuration, ZeroTrain will always return the same result and trace.

Summary

The Options object provides controlled flexibility for execution, scoring, and output formatting, while preserving deterministic and traceable decision behavior.