Fable
The archive nailed rare words to the door
In the earliest days, the archive relied solely on the old archivist's memory to find materials. When a reader asked about something, the archivist would go rummage in the storeroom based on impression. This worked when the archive was small, but as files multiplied, many papers with clear names written on them were overlooked.
One day a merchant wanted to find the "black tung wood tax deed." The old librarian first searched by the three broad categories of trade, timber, and taxation, bringing out a pile of similar files, but missed an old deed whose title directly said black tung wood. The merchant was very annoyed.
The curator first asked the old librarian to read more books and remember more relationships. But this approach became increasingly tiring and was not good at catching rare but critical terms. The second fix was to rearrange all the files by theme, but looking up names of people and places was still slow.
Later the librarian made a term card for each file. Common characters got low weight; rare words that could distinguish materials got high weight. When someone asked about "black tung wood tax deed," the term card could immediately pull out the files containing those keywords.
This approach does not understand all meanings. If readers rephrase things completely differently, it may not be flexible enough; but when the question contains clear terminology, numbers, place names and proper nouns, it is fast and stable.
The archive finally adopted it as a basic search method. The curator understood that some retrieval depends on semantic similarity, while some retrieval must first capture rare words and terms. Using sparse key markers well can quickly locate candidates in massive materials.
The old librarian also used rare cipai for the first round of screening, then handed them to close readers for judgment. Cipai can find candidates but can't guarantee each one is correct; it's responsible for speed, and final relevance still depends on the specific content.
Later the archive kept two kinds of search at the same time. When asking "contract number so-and-so of a certain year," the sparse term index was almost irreplaceable; when asking "cases similar in meaning to this matter," another method was still needed. Only by knowing the applicable boundaries can one avoid treating fast lookup as universal understanding. Later, when readers asked about rare medicine names, ship numbers, or case file numbers, the term index could always pull the material onto the table first. It does not pretend to understand all meanings, but when clear clues appear it gives a reliable starting point. Such a starting point is not glamorous, yet it often determines whether there is any chance of finding the right answer later. For this reason the curator also placed it early in the search process, so that later judgment has better candidates. Only when the first step finds accurately is there room later. This step is plain, yet it often saves an entire search.