The deployment agent also wrote release documents and other platform documentation, so I built and tuned a prototype evaluator for them. Release documents follow a fully specified format, so they were checked objectively against a master document. Other documents were scored 60% on content (truthfulness, authenticity, language, accessibility and understanding) and 40% on format and chronology. Long documents moved to 65/35, because following a template faithfully on a long document makes it exhausting to read.