What a prediction market quant actually does

The job, in plain terms

Prediction markets print a price for every contract. Most tools stop there and call the price an opinion. A quant does three more things. First, build an independent forecast, so you have something to compare the market against. Second, charge every real cost of acting: the depth actually available, the venue's fee formula at your fill price, and the chance the quote moves before your second leg fills. Third, score the forecasts after markets settle, because a model that never gets graded is a superstition with a user interface.

What that looks like in eagL

• Crypto strikes are priced from a geometric Brownian motion model fit to stored minute candles, with drift and EWMA variance estimated per asset. The crypto markets page walks through the model.
• Every candidate trade is charged real order-book depth and each venue's own fee schedule before you see a number. The arbitrage page shows why this kills most apparent spreads.
• Resolved forecasts are scored with Brier score, log loss and a ten-bin calibration curve. Backtests run on a chronological 60/40 holdout, covered on the backtesting page.
• Statistics only publish after minimum samples: 100 detections, 500 quote frames, 50 paper fills, 100 settled outcomes and seven days of watching.

Why local matters

eagL runs on your Mac, against your own venue accounts. API keys live in the macOS keychain and never cross the network to any eagL server, because no eagL server exists. The build is read-only and paper-first. If you want live order placement you switch it on yourself, behind a kill switch, and it never exists for Polymarket at all.

What it will not do

It will not promise profit. A correct scan usually ends with the edge dying at the fee line, and eagL says so instead of burying it. The full method, including its limitations, is written up on the methodology page.