Example: https://pt.wikipedia.org/wiki/You_Need_to_Calm_Down
If you look at https://pt.wikipedia.org/wiki/You_Need_to_Calm_Down?action=cirrusDump you can see recommendation.image/exists|1 in weighted_tags. The page already has an image, so it shouldn't have a suggestion
The image actually has no suggestions, and in analytics_platform_eng.image_suggestions_search_index_delta for snapshot=2022-06-27 there is this line, which should have caused the flag to be removed from the search index:
ptwiki 0 6019454 recommendation.image ["__DELETE_GROUPING__"]
I suspect that the _full search index dataset (which doesn't contain deletes) was imported rather than the _delta dataset. We know that the _full dataset was being imported until recently, and we also know that old _delta snapshots contained far more deletes than they should have. Not really clear what we need to do to repair the data, but we probably need to figure out a way
edit: one way we might repair the data is to do a search for all images with suggestions, then use the api to check if they really have suggestions, and delete the hassuggestion flag from the search index if not