Most writing about prediction-market arbitrage quotes the best spread someone has seen. This article does the opposite: it takes one snapshot of the numbers Arbitrage Agent publishes on its /proof page and reads them without rounding them up. The snapshot was taken on 4 Nov 2026 at 05:38 UTC, covering the previous 24 hours. The live page changes continuously, so treat these as one dated reading, not a standing claim.
What the scanner saw
- 1,031 feed readings across 23 distinct pairs of matched Polymarket and Kalshi markets in the last 24 hours, with no hour without a feed.
- Median net edge per reading: 0.59%. The 90th percentile was 1.23% and the best reading was 3.53%.
- Median book depth: 137 contracts on the thinner leg (90th percentile 927, maximum 5,965).
- Median realizable profit per reading: $0.62. The 90th percentile was $5.67 and the maximum $28.74.
- Median time to resolution: 221.8 days. The 90th percentile was 734.4 days.
How to read the sample size
1,031 readings across 23 pairs is about 45 readings per pair. They are repeated measurements of the same pairs as prices moved, not 1,031 independent opportunities. The number of distinct pairs is the better indication of how much there was to look at. Over the last 7 days the scanner saw 60 distinct pairs. Of the 23 seen in the last 24 hours, 9 were new and 14 had been seen before, so a good part of what you see on a given day is the same pairs reappearing.
The part that changes the picture: time
A 0.59% edge sounds small but fine. Add the time to resolution and it reads differently. As a rough illustration only, since it combines two medians taken from different distributions, 0.59% earned over about 222 days is roughly 1% a year before any other cost, and the capital is locked on both platforms until the markets resolve. That is the reason capital lockup matters at least as much as the headline spread. The 90th-percentile pair resolves in about two years.
The honest result of the simulation
The /proof page also reports a 30-day simulated (dry-run) result: the engine detected opportunities on 26 of the last 30 days, and the accumulated simulated P&L was $0.43. That is a simulation, not real trading, and it is small. It says the detection runs most days; it does not say the strategy produces material income at this scale.
What this data does not say
- It does not say these readings could all have been traded. Fees, depth and the time it takes to place two orders are the gap between a reading and a fill; see how Kalshi fees are calculated.
- It does not say that matched pairs resolve identically. The risk that two markets are not the same bet is covered in resolution mismatch.
- It is one day. A different day will give different medians.
Where Arbitrage Agent fits
Arbitrage Agent matches markets across Kalshi and Polymarket and shows the result in a live feed with a step-by-step guide for both legs. You place the orders yourself, and the final check is still yours. The numbers above are the ones it publishes about itself, small edges included. See the plans or start with the Starter plan.
To try your own numbers, use the free arbitrage calculator: it takes the two prices, your stake and the days to resolution, and shows the fees, net edge and annualised return.
Related articles
- Capital Lockup and Settlement Time
- How Kalshi Fees Are Calculated
- Polymarket UMA Disputes and Resolution Mismatch
- Live numbers on /proof
Key terms: Event Contract · Resolution Risk