Data studies on how Kalshi and Polymarket markets actually behave.
We fire thousands of graded signals on Kalshi prediction markets and log every one — win or loss — into a public dataset. These studies are what that data shows about the markets themselves, including the study in which our own order-book instrumentation went against our founding thesis: after an alert, the alerted side does not measurably move. Nothing here is a pitch. Every live number is reproducible from the public results log and the downloadable CSV. The NFL study is a season replay, labelled a backtest throughout, with its full method and its uncertainty published alongside it.
One short report a month: windows settled, contracts traded, how often Up won, how far Bitcoin moved from the strike, how well the ask matched outcomes, and how our own graded feed did, losses included. Latest edition: September 2026.
Kalshi's own public settlement data, checked the way a skeptic would: every result follows the published rule, each strike equals the previous window's settlement value, and Up and Down are balanced. One real pattern: results lean against streaks, by less than the fee on a 50¢ contract.
Bitcoin · Favorites · 15,183 windowsYes, about once in 100: 99¢ asks lost 0.97% of the time, close to what the price implies. One loss erases 99 wins, the 1¢ single-contract fee takes the rest, and most losing favorites finished within $25 of the strike.
Bitcoin · Strike · 25,317 windowsA settlement equal to the strike resolves Yes. None of 25,317 windows tied, but 1 in 117 settled within $1 and 1 in 22 within $5, and in those close finishes the side that was ahead three minutes out lost 44.1% of the time.
Bitcoin · Time of day · 25,317 windowsVolume, size of moves, near-flat finishes, spreads and price accuracy for every hour and weekday. The 10 AM ET hour moves most, Saturdays are calmest, and no hour beat or missed its price by more than chance.
Bitcoin · Fees · 11,803 signalsKalshi's taker fee worked out at every price, then tested on 11,803 graded signals. A one-contract order at 99¢ pays a 1¢ fee, so its breakeven is 100%. No entry-price bucket in our record beat its own price by more than chance; the 90 to 92¢ bucket fell short of its price by 2.80 standard errors, just inside the bar for ten tests.
Bitcoin · Settlement · 18,377 windowsSettlement is a 60-second CF Benchmarks BRTI average. A Coinbase/Kraken spot read at the close disagreed with the official result on 4.37% of windows and 39.2% of windows within $5 of the strike (corrected October 5, 2026; refreshed through October 8). The Kalshi ask tracked the outcome at five and three minutes out, and weekends had nearly twice the share of knife-edge closes.
Bitcoin · Polymarket vs Kalshi · 55,334 windowsSettlement source, tie rule and fee rounding side by side, plus 55,334 officially settled windows. The venues landed Up 49.9% and 49.8% of the time, the move with two and a half minutes left decided most windows, and on Kalshi the ask already priced it in (Kalshi figures corrected October 5, 2026).
Bitcoin · 9,800+ windowsThe win rate on 15-minute BTC contracts tracks the price you pay almost exactly — and near-flat windows are a coin flip, every month, because settlement noise dominates. A rigorous look at what a high hit rate on these markets is really made of.
Our founding thesis was that football win probability moves in discrete jumps the instant a play ends while resting limit orders do not. We argued it, then we instrumented the order book around our own alerts and found a median alerted-side move of 0.0000¢ at every horizon out to five minutes. The page keeps the argument, publishes the correction, and publishes the full-season replay we used to test it, plus the timing leak that killed our first headline. Backtest, not a live record.