A million judged questions

You can try it. The post points at a public page where you submit a question and get a judgement back, with no key of your own. It is a hosted demo running on the operator's quota, so expect rate limits, and the entry copy flags the results as illustrative rather than measured.

No, it is a volume target: how many judged questions the crowd can push through the app. No accuracy, latency or cost criterion is attached to it, so hitting the count would say something about interest in the demo rather than about how well Jev decides.

The post only says the two work great together and tags both accounts; it does not document the schema, whether judgements are persisted, or how per-visitor limits work. The claim that the demo stores and judges questions through Convex is the directory's reading of that collaboration, not something the post spells out.

A judgement, and that is the point of the demo. The author is explicit that it will not answer, only judge: you get a typed pick with probabilities your code can branch on, instead of prose you have to parse and validate before you can act on it.

Wayne Sutton, who lists himself as Convex's head of community, events and startup programs, posting from his personal account. He tags both @typesafeai and @convex and calls them a good match, so the demo reads as a Convex community project rather than a TypeSafe product page.

Because it is open to the public. In a reply he says he has to add a mod feature because you know, the internet, which acknowledges that anyone can submit a question to the judged feed. Nothing states what gets filtered, how fast moderation runs, or whether abusive entries are hidden or just removed.

None is stated. The post only says let's see if we can get to a million questions, with no deadline, prize, leaderboard or accuracy target attached. Hitting the number would measure how many people showed up to submit, not how well the model judged them.