Developer API

Selected Output Columns (Derived Feature)

ZeroTrain allows you to return specific column values from the matched row (or rows). This is a derived feature of the inference process and does not impact how matching or scoring is performed.

What This Means

After a row is selected during inference, you can choose which columns to return in the result. These columns are simply retrieved from the matched row — they are not required to participate in the decision logic.

  • Does not affect inference
  • Works independently of conditions
  • Returns raw values from the matched row

Why This Matters

This is how ZeroTrain moves beyond returning just an action like Buy or Sell. It allows the model to return actionable parameters such as quantities, thresholds, routing decisions, or configuration values.

Example

Below is a sample rule set with multiple columns. Only some of these columns may be returned.

ID CallIntent SentimentScore IsVIP HasPaymentFailure Weight Action
CC-001 BillingDispute < -0.50 true true 0.9 EscalateToSupervisor
CC-002 BillingDispute null null true 0.75 RouteToTier2
CC-005 null null null null 0.2 RouteToTier1

During inference, one of these rows is selected based on matching and scoring.

Once a row is selected, you can return specific columns from that row.

Example Output Selection

Using OutputSettings, you can specify which columns to return:

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

If row CC-001 is selected, the result would include:

  • Action → EscalateToSupervisor
  • Weight → 0.9
  • HasPaymentFailure → true

Important Notes

  • Columns do not need to be part of the inference logic to be returned
  • You can return any column from the matched row
  • Column selection is controlled through OutputSettings
  • This is strictly a result selection mechanism and does not influence scoring

Summary

Selected Output Columns allow ZeroTrain to return not just decisions, but fully parameterized results. This makes the system practical for real-world execution, where decisions require additional context such as amounts, limits, routing instructions, or configuration values.