If left blank, the app uses your worker at polywatch-proxy.rttrading3682.workers.dev. It must be running the updated GridEdge worker script so it allows ESPN.
Adds a consensus spread and total across US sportsbooks. Odds are fetched at most every 8 hours per league (2 credits each) to stay inside the free 500 credits a month.
The rest of the weight goes to ESPN's FPI projection and the team scoring model. Markets are sharp, so values under 40% tend to overstate edges.
Checks each way the app can reach ESPN and shows which ones work on this device.
The one pattern that has won every week so far: college game-total Overs the model rates 52% or higher (27-12 in weeks 1–3, 8-3 live in week 4). Everything else below is for tracking only. The Results log option "Recommended bets only" tracks how this rule does from here on.
Every upcoming pick sorted by estimated hit probability. The white tick marks 52.4%, the break-even rate at standard -110 odds. The scale runs from 50% to 65%.
Upcoming games where 65% or more of Covers contest players are on one side. The bar shows the share of users (50% to 100%); it turns red when the model disagrees. Heavy public sides have not reliably won over time, so check the Results log (bet type: Covers consensus) to see whether following or fading the crowd is working this season.
Projected scores blend the market line, ESPN's FPI win probability, and each team's season scoring adjusted for opponent and injuries. Open a game to see the inputs and the injury report.
Live picks are the ones the app saved before kickoff. Backtest replays completed weeks using only what was known before each game. Units assume 1 unit risked per pick at -110, so a win pays +0.91 and a loss costs -1.