yes, you can assume that any observations in such a project are annotated as Dead.
i would just add that some observation fields (such as “Dead or Alive”) are linked to specific annotations. if you were to add the observation field “Dead or Alive” with the value “dead” to your observation, that would automatically add an “Alive or Dead = Dead” annotation. similarly, deleting that observation field would remove the annotation. Adding the annotation does not add an observation field though.
it’s also worth noting that annotations the voting mechanism. so even if someone annotates an observation as, say, “Dead”, another person could come along and downvote the annotation. if the net votes for a particular annotation value (including the original annotation) <=0, then that particular observation would be ignored when someone filters observations by that annotation value. (it would also drop out of your project.)
if doing this sort of thing, i would download all observations first, and then separately download a set filtered for each annotation value i’m interested in. then use those subsets to populate a column (or multiple columns) in the “allt” set. (this is the easiest way to make sure you’re also capturing unannotated observations, when they exist.)
but for a set that is only ~15000 observations, it’s also possible to just get the observations via the API and parse out the annotations and observation fields you’re interested in. (i have a Juptyer notebook that could be adapted for this: https://jumear.github.io/stirpy/lab?path=iNat_APIv1_get_observations.ipynb, but there are lots of other ways to get this information, too.)