invoke
abstract operator fun invoke(input: Input, output: Output, gold: InputLabel, fullTrace: Prompt?, subgraphTraces: Map<String, List<Demonstration>>): GEPAFeedbackResult
Returns the rollout score and module-level feedback.
In GEPAFeedback implementations, fullTrace typically will be used to provide feedback for optimizing the strategy instruction, while subgraphTraces will typically be used to provide feedback for optimizing subgraph instructions.
Parameters
input
Dataset input used for the rollout.
output
Agent output produced by the rollout.
gold
Expected label or reference value for input.
full
Full strategy prompt trace, when available.
subgraph
Demonstrations captured for each optimizable subgraph during the rollout.