Not a bug, AFAIK the auto-captive vote depends largely on location. I’ve worked to tag many species that never or very rarely naturalize in California so that there’s less effort wasted identifying them and tagging them as captive.
If you’re patient, you’d basically have to go through and mark a large percentage of captive observations for a particular species, and the next time the geomodels update, it may get auto-voted captive from then on.
I think it needs to be something like 70-80% of observations need to be tagged as captive in the smallest unit of a place (county, province, etc) before auto-vote is enabled.
More notes here: https://forum.inaturalist.org/t/why-are-some-taxa-automatically-captive-cultivated/15638
Make sure you’re only voting observations as captive if you have reason to believe they’re either planted and/or don’t naturalize in your area however, there can be surprises, like cork oak being naturalized in CA while also being widely planted.
Big caveat: I believe the observations need to be at species level to get auto-voted for the most part