Use these values when mapping evaluator prompt variables to live data.
Target-specific rules:
target=observation supports input, output, metadata, and tool_calls
target=experiment supports input, output, metadata, tool_calls, expected_output, and experiment_item_metadata
Source semantics:
input: the observation or experiment input payload
output: the observation or experiment output payload
metadata: the metadata object for the target. Combine with jsonPath when you need one nested field instead of the whole object.
tool_calls: the tool calls recorded on the observation, as an array of {id, name, arguments, type, index} objects in the order the model emitted them. Combine with jsonPath (for example $[*].name) to select parts of each call.
expected_output: the experiment item's expected output. Only valid for target=experiment.
experiment_item_metadata: the experiment item's metadata object. Only valid for target=experiment.
Source field used to populate a prompt variable.
Use these values when mapping evaluator prompt variables to live data.
Target-specific rules:
target=observationsupportsinput,output,metadata, andtool_callstarget=experimentsupportsinput,output,metadata,tool_calls,expected_output, andexperiment_item_metadataSource semantics:
input: the observation or experiment input payloadoutput: the observation or experiment output payloadmetadata: the metadata object for the target. Combine withjsonPathwhen you need one nested field instead of the whole object.tool_calls: the tool calls recorded on the observation, as an array of{id, name, arguments, type, index}objects in the order the model emitted them. Combine withjsonPath(for example$[*].name) to select parts of each call.expected_output: the experiment item's expected output. Only valid fortarget=experiment.experiment_item_metadata: the experiment item's metadata object. Only valid fortarget=experiment.