Content metric
Unresolved Reference Rate (URR)
Also: URR, unresolved reference rate, pronoun reference rate
A passage with high URR uses "it," "this," "they," "those," and "the company" without explicitly naming the subject. The human reader fills the antecedent from surrounding paragraphs. A language model performing extraction cannot. The fix is to repeatedly name the entity in every paragraph.
URR is straightforward to audit. Count every "it," "this," "they," "those," and pronoun in the passage. Check whether each has an explicit, in passage antecedent. Divide unresolved references by total references. Multiply by 100. The result is URR. Target below 5%.