But if you could teach the AI to evaluate the posts and their comments in the sense mentioned, the upper limit of beneficiaries in Python would only be a minor problem.

Ugh... I started this from the perspective of handling new posts, so there would be no replies or resteems to worry about. I was happily ignoring those. But, you're right that this would be a good way to improve it for the historical posts. I'd probably need a better LLM model and a machine learning step in order to really do that well, but I'll start giving it some thought. I can probably make some progress by passing the replies through the free LLM. I am also starting to think that I might need to switch from a screening-only perspective to a screening + scoring approach for the pre-AI checks - as with the trending scores.

Thanks for the feedback!