So this is probably restating some of what was said above, so consider yourselves “quoted” :-)
This is may be nitpicking the terms, but I’d suggest that an iNat “record” that includes multiple species in different pix isn’t “an” observation. It’s a bunch of observations all glued together. This is, maybe(?) where the term “observation” as in “something I saw in the field” flows into “observation” as in a record in iNat representing the thing I saw in the field. I would agree that the thought/concept/experience/whatever of the observer in the field belongs to them. I’m not as convinced wrt the iNat record, at least once the observer has left the community and is no longer able to curate their data.
As others have noted, the point of iNat (or at least one point) is to make useful information available. Multi-species records do not achieve this. So if it’s a choice between maintaining a (useless) original intact record and changing it with clear documentation and traceability back to the original, thereby making it potentially useful, I’m for the latter.
There are, as mentioned earlier, caveats. A record with poor or missing date and place will not be improved by splitting, so those are probably washouts. Observers who opted out of community ID require some thought. And multirecord observers who are still active (or have been active in the past, say, 2(?) years) should be encouraged to split but not forced. But once an observer has left, I think the observations are fair game. After all, they posted their info to contribute to the community.
I should note that I have spent over 30 years working in clinical research data, which is highly regulated and extremely conservative, so I make these statement with much careful thought and some trepidation lol.
And I’m also very wordy - sorry for the tomes!