A different prediction

Twelve months ago, prediction markets were a niche interest for a few thousand crypto-native traders and election-cycle news junkies. Today they're being called one of the fastest-growing corners of finance, and the numbers back that up. But the story underneath the growth curve is more interesting than the curve itself.

The growth is real, and it's recent

Pew Research Center's analysis of trading data from The Block found combined monthly global volume on Kalshi and Polymarket rose from under $5 billion in September 2025 to roughly $24 billion in April 2026, an increase of roughly 5x in about seven months (Pew Research, May 2026). For scale, Pew notes that's already in the neighborhood of the ~$14 billion a month wagered through legal U.S. sportsbooks in 2025.

Full-year 2025 figures vary depending on how volume is counted: notional versus taker volume, and whether Polymarket US is included, produces meaningfully different totals across sources. Industry trackers put total 2025 sector volume anywhere from roughly $44 billion to $63 billion (Gambling Insider; Bitcoin.com News, May 2026), while other analyses using gross/notional accounting report substantially higher combined figures for the two leading platforms individually. Whichever measure you use, the direction and the magnitude of the trend are not in dispute: this went from a rounding error to a real market in about a year.

What actually drove it

This wasn't organic growth alone. A specific sequence of regulatory and institutional events cleared the runway:

• The regulatory thaw. A new CFTC chairman withdrew proposed rules that would have restricted prediction markets, and Polymarket received a CFTC no-action letter, reducing enforcement risk (TRM Labs, March 2026). Polymarket had been fined and forced out of the U.S. in 2022 for operating as an unregistered exchange; its 2025 acquisition of CFTC-licensed clearinghouse QCEX for $112 million paved the way back in (FalconX, February 2026).
• Institutional capital arrived. ICE/NYSE announced a strategic investment of up to $2 billion in Polymarket at an $8 billion valuation, the clearest institutional legitimacy signal the sector has gotten (TRM Labs). Kalshi's own valuation reportedly jumped from $11 billion to $22 billion in about six months (Morning Brew, June 2026).
• Distribution through mainstream brokerages. Robinhood's March 2025 partnership brought a prediction markets hub to its 27 million funded brokerage accounts, and Super Bowl-related volume on that integration alone exceeded $1 billion (TRM Labs). Robinhood followed that in July 2026 by launching its own layer-2 blockchain, explicitly positioning prediction markets as one of the asset classes meant to live on it (Forbes, July 2026).
• Sports became the on-ramp for a new audience. Sports event contracts made up 87% of Kalshi's March 2026 volume, and the FIFA 2026 World Cup drove roughly a 75% jump in daily sector-wide volume, with organizers noting sports specifically attracts traders who wouldn't otherwise have come to these platforms at all (Coinspot, July 2026).
• Capital followed. Prediction markets were the single most-funded category in crypto in the first half of 2026, pulling in $1.85 billion, about 26% of all funding raised across the top ten crypto categories (Coinspot).

Here's the part that gets skipped: "accurate" is a statement about averages

The industry loves to cite calibration numbers, and the headline figures are genuinely impressive. Analyses of large market samples report overall calibration in the 92-94% range, meaning that when a market says something is 30% likely, it happens close to 30% of the time, averaged across thousands of markets (WEEX, 2026; StockAlarm, January 2026). Academic work going back to the Iowa Electronic Markets literature backs this up at a structural level: markets have been shown to outperform polling averages at long time horizons, sometimes substantially (International Journal of Forecasting, 2025).

But average calibration across thousands of markets is not the same claim as "this specific bracket, on this specific market, right now, is priced correctly," and the research that looks below the aggregate finds real, persistent cracks:

• A Vanderbilt study analyzing over 2,500 markets and $2.4 billion in volume from the 2024 election cycle found directional accuracy varied enormously by platform: PredictIt at 93%, Kalshi at 78%, and Polymarket at 67%, despite Polymarket carrying by far the most volume (Clinton & Huang, 2025, via arXiv). The authors note this is consistent with the idea that concentrated positions on higher-volume, uncapped platforms can dominate visible prices in ways that don't track truth as well as smaller, capped markets do.
• A December 2025 study using 478 million Polymarket trades found no general longshot bias, but did find a documented tendency for traders to overtrade the default "Yes" option, a real, measurable distortion sitting inside the same platform the aggregate-calibration numbers are drawn from (Reichenbach & Walther, SSRN 2025).
• Weather markets specifically have already become one of the fastest-growing segments in the space, with Kalshi alone processing over $200 million in that category (TradeAlgo), and, as anyone who's actually modeled temperature forecast uncertainty knows, a naive assumption that these markets are symmetric or well-calibrated by default doesn't hold once you check it against an actual ensemble forecast or climatological distribution.

None of this contradicts the growth story. It sharpens it: the more volume and the more categories pile in, the more the "the market is basically always right" heuristic breaks down into something closer to "the market is right on average, across huge samples, in ways that say very little about the specific bracket in front of you."

The gap nobody's building for

The tooling that's grown up around this explosion is heavily weighted toward following: whale trackers, copy-trading bots, cross-platform arbitrage scanners, news-driven signal feeds. Almost none of it asks the more basic question: does this market's price actually match what a properly built probability model says it should be? The one notable exception, a public calibration-scoring site, is backward-looking, grading platforms after markets resolve, not helping a trader check a live, open bracket before they take a position.

That's the corner that's actually underserved: not more speed, not more copy-trading, but genuine probabilistic determinism: building the distribution a market should have, from ensemble forecasts, historical base rates, or count-process models, and holding the crowd's price up against it before you decide whether to trust it. The volume proves people want to trade this stuff. The literature proves the aggregate is trustworthy but the specifics aren't automatically. Nobody's built the tool that lives in that gap yet.

Building a platform worth using

What we actually care about, longer-term, is making this a platform serious traders keep coming back to, and that comes from real product depth, not from a headline number. A couple of directions we're looking at:

• Better tooling around the trade. Growing beyond read-only analytics toward a genuinely better trading experience, for example surfacing and helping manage order types the underlying platforms don't support well, like stop-loss and take-profit. Polymarket's own interface has been described by traders as offering "no advanced order types, no automation, and no portfolio intelligence" next to third-party terminals, so the gap is real. Anything that touches actual order execution would be built only within the appropriate regulatory structure; we treat that as a precondition, not an afterthought.
• Deeper analytics as a tiered product. The distribution modeling, resolution-risk scoring, and market organization tooling already built are naturally suited to a subscription or usage-tiered model: the core discovery and organization layer free or low-cost, the calibration and modeling layer as the paid tier.

The path to retail viability

This is the highest-leverage strategic decision in the whole plan, and it's worth being precise about what it is: a product-architecture choice, not a financial claim of any kind. It points at a market that's already enormous and largely untouched by anyone building in this space today.

Kalshi alone controls the large majority of the regulated US prediction market, and its users fund accounts in dollars and never touch a wallet, a token, or gas fees at any point. Even Polymarket, the more crypto-native of the two giants, is carrying real friction that keeps its own users from ever becoming "crypto users" in any deeper sense: traders report wanting to deposit directly from a bank account and being unable to, having to buy USDC first and eat a conversion fee, and finding withdrawals a multi-step, multi-hour process next to a traditional broker's two-click cash-out. That friction isn't a crypto-native complaint. It's the exact reason the mass of people already trading these markets stay mass-market rather than becoming a crypto audience, and it means nearly every tool built for this space so far has been built for the smaller, already-onboarded slice of that userbase, not the much larger one sitting right next to it.

That's the opportunity: everyone currently building in this ecosystem (the whale trackers, the copy-trading bots, the arbitrage terminals) requires a wallet, requires gas, requires the user to already be a crypto participant before they can even open the tool. None of them are built for the millions of people already trading Kalshi and Polymarket in plain dollars through a regular account. If the core product here (distribution modeling, market organization, discovery) is built to work through an ordinary account and ordinary billing, with no wallet required to browse, organize, or analyze a single market, the addressable audience isn't the few thousand people comfortable with self-custody. It's the entire existing base of both platforms, most of whom have never needed a crypto wallet for anything else in their lives and have no reason to start now just to use an analytics layer.

Today, deliberately, the product isn't there yet, and that's by design. The current MVP is wallet-gated: to use it right now, you connect a wallet. That's a choice about who we're launching to first, not the end state of the product. Gating the first release this way scopes the initial launch to a focused group of early supporters, people already comfortable in crypto, close enough to the thesis to give real feedback, and willing to back the build before it's fully proven, rather than opening a half-finished product to a mass audience it isn't ready to serve well yet. The wallet gate is the launch mechanism for that early-supporter phase, nothing more. The mass-market, no-wallet version described above is the direction the architecture is being built toward, and lifting the gate once the product is ready is a planned step, not a rewrite.