A ship-caulker records a long take that includes the phrases “oakum, then pitch.” One annotator marks when each phrase starts; another writes a single mood note on the whole file. Which annotation helps the speech recognizer learn from the clip?

On a long ship-caulker take containing phrases such as “oakum, then pitch,” time-aligned words that mark when each spoken phrase occurs help the speech recognizer learn from the clip. An unanchored mood note covering the entire recording lacks timing. A gallery caption of the caulking iron with no timestamps is not speech alignment. A Kubernetes scale setting is deployment config, not transcript timing. Exam caveat: time alignment is the annotation that ties words to audio positions. Operational check: mark phrase start (and end) times on the civic recording, bind them to transcript tokens, and verify a training example uses those alignments rather than a single mood blurb. Require time stamps on oakum/pitch phrases in the caulker annotation UI. Time-aligned words teach recognizers when “oakum, then pitch” occurs; mood notes and Helm scales do not. Prefer time-aligned word marks over shift-long mood notes whenever caulker audio will supervise a recognizer. Close q140 with a municipal evidence note that records why the correct mechanism fits this civic multimodal scenario, why each of the three distractors fails by a distinct concept, and which operational check was observed in logs or dataset cards before promotion. File that note with the run packet so later maintenance cannot silently revive a rejected framing.