I think, again, this is partly experience of another taxa and another location perhaps.
In Diptera, if I do not do this, the IDs which are most likely to be accurate languish in obscurity, the AI cannot learn from them and the GBIF dataset does not include them…
Meanwhile incorrect blind agreements made by users with limited or no experience… to original IDs made by identifiers with limited or no experience reign supreme… which increases the inaccuracy of the AI… and pollutes the final dataset…discouraging others with expertise from joining … increasing the inaccuracy, in a vicious cycle…
In any case though, I would also rather see another way of tackling it. Its totally counterintuitive - thats why I started this thread…and the parallel one about gamifying accuracy…