Fable
The director no longer laid the same answer across the whole table
A researcher asked the archive: which customers will churn because of a failed upgrade? The retrieval system quickly pulled ten documents, all with high scores. But the curator frowned. Eight of the ten came from the same incident post-mortem, just with different titles and summaries. They were all relevant, but once they filled the table, there was no room for other risk signals. The craftsman first deleted duplicate documents. But the problem wasn't solved: the customer service records, tickets, meeting notes and patch notes from the same incident weren't exact duplicates, yet they all told the same angle. Someone suggested sorting by score alone. The higher the score, the higher it ranked. As a result, the evidence the agency read became increasingly concentrated, and the answers were confident but missed secondary causes like insufficient training, permission configuration and customer communication.
Later the curator changed the table-setting rule. Every time a new piece of material was added, two questions had to be asked: how relevant is it to the question; and how different is it from the material already on the table.
On first application, the whole setup did not put all ten records of the same incident into context, but retained one retrospective, one work order, one customer email, one version note and one renewal risk record.
The agency's answer changed. It no longer just said 'upgrade failure caused churn'—it could break out technical failures, communication cadence, compensation plans, and client internal approval delays.
The curator reminded the team: diversity is not for looking good, but for allowing limited context to cover more complementary evidence. Relevant but repetitive material steals the positions that could hold truly incremental information.
From then on, the archive desk no longer pursued making every sheet look most like the question, but rather making the whole most helpful for answering the question.
When the new rule was first introduced, the researcher was not used to it. He was accustomed to taking away all the highest-scoring papers, thinking that a slightly lower score meant poor quality. The director had him paste ten papers on the wall by reason, and only then did he see that high-scoring papers often merely restated each other.
Later the archive left a note for each paper selection: which angle this one supplemented, and which duplicate material it replaced. The limited desktop was managed as a scarce resource, and answers expanded from a single incident to a complete causal chain.