How ArbMonster measures

Every number on this site comes from prices we recorded ourselves — 8,344,054 order-book observations since Aug 9, 2026 — and every rule that turns a price into a card is written down here.

1. Sources

Kalshi: authenticated websocket order-book feed for every confirmed market, full depth both sides. Polymarket: public CLOB websocket, YES and NO tokens, full depth. Sportsbooks: pre-game moneyline, spread and total lines for 20+ US books via an odds aggregator, refreshed every 15 minutes. In-play lines are deliberately not scanned — they go stale in seconds and a 15-minute cadence would mislead.

2. Matching the same question across venues

Kalshi lists roughly 100,000 markets and Polymarket thousands. A matcher proposes candidate pairs from titles, close dates and rules, then a language-model reviewer judges whether the two contracts settle identically and returns a confidence. Only confirmed pairs are scanned, the confidence is printed on every card, and pairs whose game dates conflict are rejected automatically. Sportsbook markets are matched deterministically by event, market type and exact line; a −1.5 at one book is never paired with a −2 at another.

3. Sizing at real depth

For prediction-market legs the engine walks the order book level by level up to the stake and uses the volume-weighted average price, not the top quote. Displayed size is haircut by 50% in the simulator to account for other takers. Sportsbook depth is not visible, so book legs are marked “depth: unknown” and sized at a conservative reference stake. If the two legs cannot be sized to the same payout, the smaller one governs.

4. Fees

Kalshi: 7% × contracts × price × (1 − price) per taker fill, rounded up to the cent, with series-specific overrides. Polymarket: shares × category rate × price × (1 − price) on taker fills, with the category read from the market. Sportsbooks: no explicit fee — the margin is inside the odds and is what the vig pages measure. Net edge is (payout − cost − fees) / (cost + fees). The gross edge is shown only for comparison and is never a headline.

5. Freshness and exclusions

Prediction-market cards re-evaluate on every book update and are dropped if not re-detected within five minutes. Sportsbook cards survive only if the most recent completed scan re-found them. Edges above 15% are excluded from every page as line mismatches. Cards older than the freshness window are removed rather than shown with a warning.

6. Honesty metrics

Each detection is re-priced 5, 30 and 60 seconds later against the then-current books and the surviving edge is stored and displayed. A 24/7 paper-trading simulator applies the same fill model plus latency, legging and a 10%-a-year annualized hurdle; its results are simulated and are never presented as returns.

7. What the site is not

ArbMonster never places orders, holds funds or connects to a venue account. It does not give betting or investment advice. Venue availability depends on your state, both venues' rules can differ in edge cases, and every price should be verified on the venue before acting.

Questions about the method: contact@arbmonster.com. See also about, venue costs, glossary.