eagL / arbitrage after fees

Why prediction-market arbitrage usually dies at the fee line

Scanners love to show you a YES on one venue and a NO on another that sum to less than a dollar. Most of those spreads are not money. They are the gap between a screen price and an executable price, and this page walks through where the edge actually goes.

A worked example

Take a cross-venue pair with a headline spread of +4.20%. That number assumes you fill both legs at the displayed price, at any size, for free, simultaneously. Charge reality and it moves fast: walking the book to a realistic size shifts the entry, both venues take their fee at your fill VWAP, and one venue takes its cut of profit at settlement. In a typical case the +4.20% headline becomes −5.08% of costs against the spread, for an executable edge of −0.88%. The correct decision is reject, and a scanner that never says reject is not scanning.

The three places the edge dies

Depth. Displayed prices hold for a handful of contracts. eagL walks the ladder level by level and grows the size only while the next contract still pays.
Fees. Each venue charges its own formula: Kalshi's schedule rounded up to the cent, PredictIt's cut of profit at settlement, Coinbase at 120 basis points taker. A flat haircut gets all three wrong.
Time. Two legs never fill in the same instant. eagL replays its stored quotes thirty minutes forward from detection, aligning venues only where samples land within sixty seconds, and counts the pair as takeable only if it still cleared costs thirty seconds later.

What survives

Occasionally something does clear all three, usually small, usually briefly. eagL reports it with the depth it was reachable at and the evidence trail behind it. One honest caveat: quote history is top of book, so the replay proves the price held, not that the full size sat there. We would rather state that limit than imply precision the data cannot support.

Why matching matters as much as math

The fastest way to fake an arbitrage is to pair two contracts that are not the same event. eagL matches titles with IDF weighting behind a numeric gate, so a $100k Bitcoin strike never pairs with a $150k one, and play-money venues are dropped before matching begins. The full pipeline is on the methodology page, and the validation regime behind these claims is on the backtesting page.