I have been working on a project to collate species occurrence data inherent from unpublished student theses in an integrated database (currently published in GBIF) and still working on a systematic protocol of data validation. Expert review is really subjective and I got many findings that said "expert" estimation were not always more consistent than amateurs, student, or even public enthusiasts (feel free to message me for the papers I collected regarding this), thus my team was still struggling to find a way. Our current method is just independently evaluate the scientific names through taxonomic checklists and the geographic distribution were validated through available published literature mentioning the geographic distribution of each species. We occasionally ask experts but as we are working on many understudied taxa and geographical area, there was not many around.
maybe its good to read this file " Improving wildlife data quality: guidance on data verification, validation and their application in biological recording "
I suppose it all depends on your study species. For the most part, I think experts in most fields are able to identify the species they're most knowledgeable about with relatively high accuracy, given they have enough information in the photo and geographic location to do so.
It's usually when someone gets a bit overzealous and identifies something to the species level when given minimal information and just going off of an educated guess for species most likely to be in the area.
It also depends on what the question for your study is. If you're doing an SDM for a species, you could always thin the records to about 100 and then self-verify (if you're confident in your abilities to do so). You could see if the species occurrence data has any corresponding NCBI molecular data and use DNA to verify species.
If you're using a dataset of 1000 + (or some other number where it isn't feasible to self verify each account) from inaturalist, you could query the data with >3 verified ID agreements with no "maverick" or disputed IDs. The likelihood of obtaining false positives should decrease with user agreement on a species identification.