Only collocations with a %[frequency|frequency]% higher than the limit will be included in the result. The auto setting will select the limit based on the size of the corpus to filter out noise and linguistically irrelevant items.