We ran the same Kalshi rule on paper and with real money. The misses were the winners
69 real orders against their paper twins: about half the paper number never arrived, and the gap was not the price.
At 5:43 pm on 28 October our bot sent a real order: buy YES at 86 cents, one contract. Kalshi answered: filled zero. On paper, the same trade filled and won 15 cents.
One miss proves nothing. So for two weeks we ran one rule both ways at once, on Kalshi's Bitcoin 15 minute markets: with real money, one contract per order, and on paper, the way every backtest does it.
The real record
- 69 real orders, 14 to 28 October. 45 filled, 24 missed, so 35% came back with nothing.
- The filled trades made +4.56 cents a contract after fees, +$2.05 in total, from Kalshi's own fill records. 34 of the 45 won.
The same decisions on paper
On 53 decisions, both versions fired on the same market. Counting a miss as zero, real money kept only about half of what paper claimed.
The obvious explanation is that real money pays a worse price. That is mostly not it. On the 36 trades where both filled on the same side, the two results were within a cent of each other, and both won the same 29. The median price difference was 1 cent.
Where the rest went
On the 15 decisions where the real order missed, paper booked a fill every time, and 13 of the 15 won.
That is the finding. Paper is close on price and wrong about which trades you get. The offer disappears precisely when the market agrees with you, so the trades you miss are not a random sample. They are the good ones.
Speed helps, up to a point. When we replayed 39 real signals against our recorded order book, the offer was still there 79% of the time at 30 milliseconds, 59% at 650 milliseconds and 36% at 2 seconds.
What we changed
- The live rule gets judged on real fills, not paper ones.
- We report the fill rate next to the cents, because cents on trades you never got are not money.
- We recheck paper fills against the real trade tape at every verdict. When we first ran that check, only 31 of 63 "confirmed" paper fills held up.
The live rule is still running toward its verdict at 60 settled real trades, pass or fail, on the bar written before the first order. If you test your own rules, the most useful thing you can take from this is simple: treat a missed order as lost edge, and ask which trades your backtest would really have got.
Members get the appendix: the exact paper and real numbers, all 53 decisions trade by trade, and the full breakdown.
Want to run rules like this yourself? Try Ping free for 7 days, no card needed. It counts a paper fill only when a real trade proves it could have filled.
Where this comes from: our own order records and Kalshi's fill records for the live rule, 14 to 28 October 2026, the paper lane's log on the same markets, our recorded order book for the speed replay, and the paper regrade of 14 October. Results are our own computation. Nothing here is advice. It is the record.