AI Trend Tracker
A listening engine that finds the story before it is a market, then drafts the question and the rule that settles it.
Six ingestors pull Spanish-language posts, videos, threads and press off X, Reddit, YouTube, news and GDELT, each on its own cadence and its own rate limit. Duplicates are collapsed with a SimHash window before anything is scored, because the same story arrives eleven times in an hour and a naive counter reads that as eleven signals.
What survives gets clustered and scored, and anything above the threshold is written up as a proposed market: the question, the resolution rule, and the official source that will decide it. That last part is what makes the output usable — a question nobody can settle is worse than no question.
Every stage runs as its own scheduled beat, so a dead source degrades the feed instead of stopping the chain. Nothing reaches the public site automatically: proposals land in a queue and a person approves them.