Reuters reported on 17 August that researchers had identified 7,466 U.S. midterm-related markets across Kalshi, Polymarket and Polymarket US through 10 August—16 times as many as in the 2024 cycle—and that $133 million had already been wagered on the 2026 midterms. Reuters describes the wider category as spanning outcomes from politics and economics to popular culture. On 19 August, Reuters reported another sign of institutionalisation: Cantor Fitzgerald had launched prediction-market trading for institutional clients through Kalshi, while Bernstein was cited forecasting that annual sector volume could reach $1 trillion by the end of the decade.

Those figures matter, but the part I find most interesting is what happens underneath them. Once multiple venues list contracts around the same election, economic release, sporting event or public event, the data starts to look deceptively simple. A title can make two markets appear identical even when one has a different cutoff date, geography, threshold or settlement source. The opposite can also be true: two contracts with very different wording can still resolve on materially the same outcome.

Prediction markets are becoming an identity and provenance problem as much as a pricing problem.

A raw price feed is useful only after you know what the price actually refers to. If a journalist, researcher, developer or institution compares the wrong pair of contracts, the resulting chart or conclusion may look precise while being conceptually wrong.

This is the problem I have been working on with Prediction Market Radar. The service is read-only. Its Super Match Engine attempts to identify materially equivalent markets across supported venues and keeps the original source links visible so the relationship can be checked. Just as importantly, uncertain relationships should be capable of being withheld rather than forced into a match merely to increase coverage.

Jurisdiction is part of provenance

Prediction markets do not exist inside one uniform market structure. Venues use different technical models, product types and rules, and access can vary by jurisdiction. Reuters reported in July that French internet providers were ordered to block Polymarket, while Spain had temporarily blocked Polymarket and Kalshi earlier in the year. That regulatory fragmentation is another reason the source itself needs to remain visible rather than disappearing behind a normalized number.

The middle layer

My current view is that the category is moving toward a familiar infrastructure pattern. First come the venues. Then come aggregators and data feeds. As the number of venues grows, the market needs identifiers, lineage, normalization, comparison and quality controls that make the feeds intelligible to other software and to people. Prediction Market Radar is an attempt to build part of that middle layer.

There are encouraging early signs that this information layer is discoverable beyond direct outreach. Google and Bing are crawling/indexing the public site, and in neutral AI-search tests where Prediction Market Radar was not named, Microsoft Copilot and Perplexity independently surfaced the service for relevant cross-platform prediction-market questions. I regard that as early discovery traction, not evidence of established audience scale, but it is a useful signal that the category and the service are legible to search and AI systems.

Over the coming weeks I will be watching three things in particular: how quickly market inventory continues to expand; whether institutional and publisher use of prediction-market data accelerates; and which cross-platform identity problems appear repeatedly enough that they should become standard reference-data fields rather than one-off matching decisions.

If you are a journalist, researcher, platform, developer or data user working on this problem, I am happy to compare notes. I am available for remote interviews and demonstrations at Bruce@PredictionMarketRadar.com.