This seems related to a known issue of undesired search behavior concerning CJK text. In the process of creating the search index, DT (or maybe a macOS framework used by DT) cuts a text document into searchable “bits” such as words and figures. Sometimes the algorithm makes undesirable cuts, e.g. cutting off “0.1” into two separate digits. The user, unfortunately and disappointingly, has no say in how the software makes the cuts.
While the DT team has not signaled their intention to improve this cutting process, I do hope to see better documentation of spoke process and disclosure of the potential of problematic search behavior. Among other things, the issue related to CJK text should definitely be included in the manual as a warning, since it contradicts with the developers’ claim that the DEVONtechnology kernel is “completely language-independent”.