Fable
Has a map and asks for directions
In the old city there was a huge archive holding tens of thousands of documents. Twenty years ago, an old curator led a team to extract the people, events, places, and institutions mentioned in the documents one by one, and drew them into a giant relationship network hung on the wall of the reading room. This network was the museum's treasure—a hand-drawn knowledge map.
After the new director took office, he found a problem: although the map hung on the wall, when readers came to look up materials, the administrator still flipped through documents by hand, occasionally glancing up at the map to find cross-references. The process was slow and unstable—when the administrator was in a good mood he looked at the map more; when he was in a bad mood he just flipped through from memory.
The new curator decided to do one thing: not change the map itself, but change how the map is used. He designed a workflow: a reader asks a question, the administrator first finds relevant nodes on the map, spreads along edges to related nodes, extracts the document fragments corresponding to those nodes, and synthesizes them into an answer. This workflow was written into an operations manual and placed on every administrator's desk.
The effect was immediate. For the same questions, the quality of the answers became much more stable. The map had not changed; the change happened after the map and document retrieval were systematically connected.
One day a scholar came to visit, saw the map on the wall and asked the new curator: are you selling this map? The new curator said: we don't sell maps; what we sell is the method of answering questions accurately by relying on this map. The map is the infrastructure we run queries on, but what we ultimately deliver is an answer that synthesizes multiple document fragments.
The scholar then asked: If I copy this map and take it back, will I get the same result? The new curator shook his head: The map is only data; the process is the capability. If you copy the map and take it back, you still need to design the same retrieval process, fragment integration method and quality-check rules, otherwise the map is just a pretty picture.
Later, when the archive introduced its services to the outside world, it never showed that map on its own. What it showed was: a question goes in, a verifiable answer comes out. The map was inside, but no longer existed on its own.
Later the curator treated the graph as a map and the question-answering process as a person setting out to find the way. However detailed the map, someone still has to walk it according to the question.