As has been pointed out before, arbitrarily assigning an accuracy value does not necessarily provide better information than no accuracy value at all.
e.g.:
https://forum.inaturalist.org/t/having-an-accuracy-radius-on-your-location-can-be-important-to-some-data-users/55283/
https://forum.inaturalist.org/t/missing-location-accuracy-data/38360/
I went through the few dozen of my observations that didn’t have this value, and for those where I could evaluate (e.g. next to trails and other features visible in the imagery) the locations were very good, well within a radius of 10 m which I added as an accuracy value. In other cases of observations not as near to any fine-scaled landmark, I put 30 m or 50 m to be conservative. I’m happy that these radii have a very high chance of including the true point, and I think it’s better than not putting an accuracy value. Anyone adding such values would do well to check a subset of the locations that they can independently verify, based on memory of where the photos were taken, and include an accuracy value that corresponds.
Must have positional accuracy ≤ 50 km - for CV / Geomodel https://www.inaturalist.org/posts/115962-new-computer-vision-model-with-over-2-500-new-taxa
https://www.inaturalist.org/observations?acc=false&order_by=observed_on&subview=table&verifiable=any
66 million obs excluded
?
Whether those observations are excluded or not depends on whether " ≤ 50 km " includes values of zero (i.e. not assigned). I read that statement as indicating only that observations with very large accuracy values are included; it does not make any specific claim about how a lack of accuracy value is handled.
I think there are more useful things people could be worrying about than whether users have assigned an accuracy value, given that this value may mean completely different things depending on how it was determined (the value may have been calculated by the device, the user may have guessed, they may have arbitrarily assigned it, they may have chosen a large accuracy value as a way of obscuring the location, etc.).
I doubt any researcher looking at data is going to assume that an accuracy value of zero means that the organism was seen at that exact location down to the last decimal point. Barring situations where the habitat and the pin clearly don’t match (e.g. a photo of a fish in a river but a pin in the middle of a field), in most cases they will probably assess the likelihood that it was seen within some reasonable distance of the point on the map (say, a few hundred meters) and decide whether that is plausible or not.
the most recent comment from loarie
positional accuracy unknown are included - this query shows obs with accuracy above 50k along which would be excluded from the geomodel
https://www.inaturalist.org/observations?acc_above=50000
Using that link - I corrected 3 of my own obs. No Cape Peninsula obs are excluded (well done us)