I completely agree with what has already been said. The badge allocation process must be thorough and reliable because CGT risks your credibility and reputation. Putting the quality seal without verifying that it is true, is like issuing a certificate without knowing its content, or signing a contract without having read it (stupid). Automated allocation can be circumvented, as is already the case with model scoring (models with 1 presentation image repeated 100 times, model description with entire articles from Wikipedia ...). This causes it to completely lose its value. It is useless and will also negatively affect We already have several of those useless badges (rigged, animated)
Even though automated tools are used for some checks, human intervention is required to check that everything is correct. This does not have to take up too much time per model, for an expert eye, (good modelers have a clear, consistent style, with clean meshes ...). Nor should it be applied to the million models (only in free models, more than 225,000 would be eliminated)
The evaluation criteria should refer exclusively to the model: quality, presentation and price (views, votes, comments ... are also very manipulated by the clone armies that many users have). Mainly the quality of the model should be evaluated, but if you want the highest quality you should also have a good presentation of the model: detailed description, complete model details, clear images of the mesh, UV map, textures, armature, poses ..., (the most published models lack these details.) They must have a native format, with software updated to the date of publication (many of the models that are published today, do so with software from 10 years ago, or without native format). And also another very important point is that they must have a fair price (quality is not cheap, because it involves many hours of dedication). There are many models that do not meet any of these criteria, so by this point, the number of models to examine would also drop considerably.
The quality of the model unquestionably begins with the quality of the mesh, (100% quads manifold meshes, without intersecting or distorted faces, without poles of more than 5 edges, with loops appropriate to the shape of the model, UV map without overlapping and not distorted, original textures of standard size ...) All these details cannot be falsified and are easily quantifiable (any program includes tools to verify this data, and it can be automated).
And above all, I would avoid certifying: stolen models, models made with pirated software, counterfeits, printable models that are not, those that violate copyrights, trademarks, patents, or any other law, because doing so is like affirming that you know about these violations and that you agree (something that can exploit in your face), because a certification is a guarantee, in this case of quality and is also binding. Many buyers may be deceived by not receiving the expected quality and this could lead to returns and the loss of quality seeking customers (which are the most rare). At this point, a large number of models, not suitable for certification, would be eliminated.
It is a good opportunity, so that those who sell their models for very low of their real value, are encouraged to upload it. That those who make bad presentations and do not show the mesh, textures ... are encouraged to do so and to reward those who are original and really do things well.
Do it right or don't do it (for now you haven't done very well). If you do it right you will attract quality buyers, willing to pay more, for authentic, quality models, but if you do it wrong, you will lose credibility and time (time that you could use to solve real problems already known)