EDJE isn't a hunch. It's a four-stage pipeline: ingest live Kalshi and Polymarket feeds, enrich with primary-source data, score with calibrated ML, and route the user to the best venue. Below is the method, in plain English.
Each live Kalshi or Polymarket market passes through the same pipeline. The model updates probabilities hourly and alerts users when a market moves into the +5-edge-or-better band.
Kalshi + Polymarket public APIs. Normalized into unified market objects: outcomes, prices, volume, resolution date, source.
Weather (NWS), injuries (ESPN), polls (538), Fed calendar (FOMC), demographics, line movement, sharp vs. public.
Monte Carlo field simulator (20K sims) + Bayesian updating + base-rate calibration against 10 yr of historical resolutions.
User routes to Kalshi or Polymarket via EDJE referral. We earn a fee. User gets the edge. The sharpest wager wins.
Every EDJE market probability starts from a historical base rate and gets adjusted by multiplicative factors. Each adjustment is tied to a backtested sample (minimum n = 18). The output: a calibrated probability that is systematically compared against the Kalshi or Polymarket price. The delta is the edge.
Every market type gets its own enrichment pipeline. Sports needs weather and injuries. Politics needs polling aggregation and demographic shift data. Fed decisions need the dot plot, inflation prints, and labor market reads. We don't invent data. We assemble it from primary sources.
EDJE publishes a quarterly backtest of every market it scored. Over the last 90 days we've resolved 148 markets with a 73.2% hit rate on +5-edge-or-higher recommendations and a Sharpe ratio of 2.41. The track record isn't curated. If a call missed, it's in the backtest.