In response to a programmatic interaction, respective representations of items of an initial result set are presented to an item consumer. One or more result refinement iterations are then conducted. In a given iteration, one or more feedback indicators with respect to one or more items are identified, a machine learning model is trained using at least the feedback indicators to generate respective result set candidacy metrics for at least some items, and the metrics are then used to transmit additional items for presentation to the item consumer.