A/B testing of a YouTube video essentially broke a promise that a YouTube public video, once published, becomes a named "work" that can only be retracted but not tampered. It renders the bibliography quotation inaccurate in referencing the name, and is destructive in building YouTube parts of the history of our era.

Shugetsu Ideology - Fifth Principle:
"All pasting present are parts of the irreversible history. Tampering is the most underestimated crime."

@elfile4138 All monetized YouTube channels are less about creating "work" but more like Media companies and proprietorships creating narratives. It's corporate history at best and we all know how biased and malleable these history is. ​
@yume "work" here is not about intention, as long as it can be used as copyright strike/receiving copyright strike, it is a "work" in the terms of copyright law. And although YouTube never had the desire to be carrying history, it has undeniably become so, as it continues to host countless videos that are definitive to the modern media and culture landscape. If corporate ESG is still of stake, then preserving history fait accompli is undeniably parts of it.
@elfile4138 I don't think you can A/B test the video (the "work") itself on YouTube? You can only A/B test metadata like titles and thumbnails which have a much higher bar to be considered protected work.

Regardless, copyright is a legal right and as far as my non-legal cat brain understands confers zero duty on the holder. There are limitations to your rights but I don't think being ethical about your enforcement methods is a limitation.

YouTube has always been a platform "for creators" which means right to be forgotten, etc. Hence they took out the dislike button, allow one to delete and reupload videos all they want, etc. Unless I have an ongoing legal case of subpoena I don't have a duty to keep record of what I said. I believe that's the legal part of it, this way YouTube keeps itself out of privacy/defamation lawsuits while making it a petri dish for malleable content with questionable COI and closed loop "optimization" pipelines.