A permitting coach shows that dumping an entire case history into a prompt can confuse the model more than concise relevant excerpts. What basic is taught?

A permitting coach showing that dumping an entire case history confuses the model more than concise relevant excerpts teaches relevance and concise context. Generative models have finite effective attention; irrelevant emails, duplicate inspections, and tangential complaints add noise that can bury the operative code section or current deficiency list. Better prompts select the passages that matter for the immediate coaching question. Claiming that relevance and concise context mean fitting excerpts to token limits into prompts misses the point—conciseness is about relevance selection, not a token limit that can preserve noise while cutting controlling facts. Claiming that relevance and concise context mean adding surrounding clause text into prompts contradicts the finding itself—adding surrounding text reintroduces the noise that buries the operative code section and confuses the model. Claiming that relevance and concise context mean ranking excerpts by recency into prompts confuses recency with relevance—the controlling code excerpt or deficiency may be older, and what matters is selecting the passages that answer the immediate question. Exam caveat: concise does not mean incomplete—omit noise, keep controlling facts (permit type, cited code sections, open corrections). Operational check: teach staff a "context diet": paste the current notice of violation, relevant code excerpt, and the specific applicant question; exclude decade-old email chains unless they change the legal posture; and verify coach outputs against the cited code.