I do not start with a price prediction. I start with the event, its original source, and its timestamp. A new listing, a move on a prediction market, and a promotional post are different signals. They should not carry the same weight.
Ordinary code assigns the first score. It checks the source, event type, freshness, and available market data. Only signals scoring 60 or higher reach Claude and Codex. The two models work independently and do not see each other's conclusions.
Yana's score is not a promise of profit. It answers a more useful question: how much attention does this event deserve right now, and what still needs to be verified?