been watching prediction markets for a while and kept noticing Kalshi/Polymarket showing completely different numbers than Vegas. built something to catch these systematically. pulls Kalshi and Polymarket data, then fires off requests to like 36 other sources at once - sportsbook APIs, polling sites, Reddit for sentiment, news stuff, whatever. when the numbers don't line up it flags it. tested it last night on the Paul vs Joshua boxing match. Kalshi had Paul at 86% to win. checked every Vegas book and they all had Joshua at -1200 which is 92% implied. complete opposite direction. flagged it immediately. Joshua knocked him out in the 6th round. technical stuff:

using Python with async calls cause hitting 30+ APIs at once tried GPT-4 for sentiment analysis initially but it was way too slow so built a custom thing runs on Poe so I don't have to deal with hosting costs takes about 2 minutes per analysis

questions I have:

is one successful call actual validation or did I just get lucky? what other data sources should I be checking? thinking about adding order flow tracking has anyone else built something similar for prediction markets? how fast do these inefficiencies typically get arbitraged away?

honestly wasn't sure if this would work at all. feedback appreciated especially on the methodology side.