What is the first thing to check when a bootstrapped re...

Before assuming the model is too small or the data volume is insufficient, investigate the quality of the few-shot exemplars. If the examples pulled from a public IR dataset are poor, the model will imitate low-quality patterns. Checking whether the source material is actually high-quality is the most critical first step in debugging an underperforming retriever.