Uibang Seongdo contains prescription records that share a name but differ in their source texts and ingredient wording. Merging those records without preserving their sources would make the list shorter, but a reader would have to search again to find out which text a detail came from. I kept the records separate and built a view where readers can compare their contents, then open the original source.
Searching for Cheukbaeksan (측백산), for example, brings up two records that can be viewed side by side. The comparison shows which ingredient names appear in both and which appear in only one. Its similarity score is calculated from sets of ingredient names normalised for comparison. It does not measure similarity in dosage or therapeutic effects. Keeping the names, contents and sources visible helps explain what the number can tell a reader. Compare the two records
In a talk about AI-generated work, a speaker used meeting notes to illustrate a problem with measuring efficiency. The person writing the notes might finish sooner, while several recipients spend more time reading and checking the result. Faster production alone would not show that the team’s work had improved. The same question can be asked of the comparison view in Uibang Seongdo. Watch the relevant section
A summary can create a similar burden. A conclusion assembled from several records may take little time to read. But before using it in a piece of work, a reader may need to know which records were selected, what was treated as equivalent, and what was left out. Without that information, they may have to go through the material again from the beginning.
Putting every source document on the first screen would not solve this. In Uibang Seongdo, readers can look at the comparison first and open the source details when they need them. A concise explanation can still provide that route. Someone looking for an overview may stop at the visible differences; someone using those differences to make a judgment needs access to the calculation method and the original records.
I have not measured whether this design reduces the time spent checking sources. Having a comparison view and source links does not establish that a new reader will understand how to move between them. To decide whether the screen needs more information or a clearer explanation, I would need to know where people spend time as they find differences between records and check the sources.
That is also the process I want to examine when judging whether an AI summary is useful for work. Can a reader follow a question back to the relevant evidence? Can they check the conditions and exceptions in the original? What needs to remain in a concise result depends on what its reader has to do next.
Related work: Why I kept same-name records separate in Uibang Seongdo