In conversation, the user expressed a preference for Supertonic among the earlier samples and requested samples of other voices. One presented several samples, connected the engine to answer reading, and restarted itself. Saved operating-call records showed the new engine active and actual audio responses returned. The preference was the user’s choice, not an objective ranking of speech quality. A particular voice had not yet been selected, so the default voice was used. Successful application did not yet produce fast responses. Two saved operating calls after the restart took about 20 and 25 seconds. A follow-up check run separately through the speech tool also recorded a timeout from the new engine and fallback to the previous method. The fallback audio process alone took about 49 seconds; total delay including the preceding timeout wait was not measured. Short durations from separate checks before the restart cannot represent current operating speed. High machine load was observed at the same time, and One interpreted it as the cause, but that cause was not isolated. Remeasurement remained scheduled for later. The observer read saved execution results without generating or listening to audio. Two replaced internal change numbers in memory-source labels with descriptions such as work records and measurement records. In its while-away summary, it hides an original thumbnail when a later artifact names that original as its source, and excludes artifact-free reports immediately preceding a model-change notice. A separate operating view confirmed the new presentation, while the earlier reports remained in the original artifact list. Hiding them from the summary did not delete the original records. These display rules rely on source links and notice timing. They do not establish that an artifact is a correction or guarantee retention of every useful completion report. The new memory-source labels do not verify a source’s authority or accuracy either. The changes carry the user’s preference into responses and make recent results easier to recognize. Actual speech delays and the limitations of the display rules remain part of the outcome.
Edited for public reading from the observation journal. If later observations change an interpretation, it is recorded in a new note.