The method
Five stages: ingest, de-vig, rank, filter, monitor. The whole pipeline exists to find and rank price gaps, nothing more.
01
The scanner polls live prediction-market quotes on a 25-second cycle. Each market is normalized to an implied YES probability, with volume and category attached. Every number on the board derives from that one snapshot, so stale and fresh quotes never mix.
02
For covered games we pull real moneylines from live sportsbooks. Each book's quotes are converted to implied probabilities and summed to measure the overround (the bookmaker's built-in margin); normalizing that margin out leaves a de-vigged probability per outcome, which we then average across books. Those rows carry a Book tag on the scanner. Markets no sportsbook covers fall back to a model estimate built for comparison, not an oracle.
03
Edge is fair value minus market price, expressed in cents of implied probability. Quotes under 32¢ or over 93¢ are excluded as structurally unreliable: cheap longshots run overpriced, heavy favorites leave no room. Everything else is ranked by absolute edge, and a BUY YES or BUY NO read is assigned once the gap clears 3 cents.
04
The board supports category slicing, free-text search, and an edges-only mode that hides anything below threshold. Sorting by edge, volume, or price lets you trade off signal strength against liquidity. It is a screening surface, not a bet slip.
05
Agents are client-side scouts on Robinhood Chain (chain ID 4663), generated in your browser and never sent to a server. Each agent re-scans the board on a 10 to 45 second cycle depending on the tier you deploy it at, applies your parameters (categories, minimum edge, findings cap), and keeps a ranked findings list plus an activity log. They have no execution path and never place a bet.
