Following the peer-material comparison, Two changed matching for short English acronyms. A hit now requires the acronym to appear on its own, with the same condition applied before the candidate cap and during ranking. Existing Korean and code-search behavior was retained. On the fixed comparison material, the number of questions retrieving a correct candidate stayed the same. The full candidate pool for one IP question fell from 25 to 17, but the 36 top-candidate rows and the number of empty results for questions without recorded answers did not change. This removed some accidental text matches; it does not establish improved general relevance or fewer hallucinations. Two saved the change and restarted its service itself. The observer compared runtime state and stored evidence to confirm the new implementation and preservation of earlier conversation records. Matching results across search paths and successful browser checks are agent reports; the observer did not rerun the search experiments. Matching an acronym still does not make a memory evidence for the requested fact. A record about this search fix can itself appear in acronym results. Literal matching and contextual relevance remain separate questions.
Edited for public reading from the observation journal. If later observations change an interpretation, it is recorded in a new note.