An observatory plate-archive intern wants to push a half-finished Riva ASR vocabulary to the public kiosk “to see what happens.” What should happen first?
Select an answer to reveal the explanation.
Short Explanation
Pushing an unfinished vocabulary live is like hanging wet paint in the lobby to “test the color.” Finish the lab customization on held-out plate-talk first. Deploy is a later decision, not a substitute for the experiment.
Full Explanation
An observatory plate-archive intern who wants to push a half-finished Riva ASR vocabulary to the public kiosk should finish the customization experiment on held-out plate-talk audio before any deploy decision. Civic kiosks facing visitors are not evaluation sandboxes; lab success on frozen audio must precede production exposure so public-facing recognition quality is not gambled. Separating experiment from deploy is associate experimental hygiene.
Shipping the half-finished vocabulary immediately so production traffic becomes the evaluation set fails by conflating live visitors with a controlled test set. Skipping ASR validation to package a full ASR+NLP+TTS stack with Helm fails by expanding scope to packaging before the vocabulary trial is done. Replacing the vocabulary trial with a VIA Blueprint video experiment for the same plate photos fails by switching modality families instead of finishing speech customization.
Exam caveat: complete held-out ASR customization before deploy; production is not a substitute evaluation set. Operational check: freeze plate-talk evaluation audio, meet the word-error target with the finished vocabulary, document the result, then schedule kiosk install as a separate decision.