Run light evaluations in Duaer

What light evaluations are

While you build a digital organization, a handful of examples is often enough to see how it behaves and to improve it. At that stage, reading the outputs side by side is usually enough. Formal scoring is not worth the setup yet.

Light evaluation runs each test row through the digital organization and writes the actual outputs back into the dataset so you can compare them to expected outputs when you have them.

How to run one

1. Create a dataset: put a few examples in a project data table or Google Sheet. Include input columns, optional expected outputs, and empty columns for actual outputs. For Sheets, add a Google Sheets credential first.
2. Wire Evaluation Trigger: add Evaluation Trigger so each run emits one row. Evaluate all runs the digital organization once per row. While wiring, set Max rows to process to 1, or execute only the trigger node, if you want a single row.
3. Connect it to the rest: use at least the input columns later in the flow. If you also have a production trigger, reshape both branches to the same JSON and merge them. See combining multiple triggers.
4. Write outputs back: after the digital organization produces the values you care about, add the Evaluation node’s Set outputs action and map those fields into the actual-output columns.
5. Run the evaluation: execute from the evaluation trigger side. Each dataset row runs once. Review outputs in the table or sheet; open Executions when you need detail.

When the set grows past a handful of rows, move to metric-based evaluation for summary scores. Concepts are in the evaluations overview.