theaiquant
Kalshi Polymarket 15-minute markets

AI quant.
Receipts first.

A trading bot that prices the 15-minute markets on Kalshi and Polymarket. Every test is locked in before the result comes in, and the losses get posted the same way as the wins.

Scoreboard

The bot vs the crowd vs the fee

Kalshi charges a fee every time you take a price. Most traders lose to it. Here is how much each side gave up for every dollar it bet this season.

Lost per $1 bet, after fees

Shorter bar is better.

The crowd (average Kalshi taker)2.2¢
The bot0.4¢
Before Kalshi's fee, the bot finished the season in profit. After the fee it is slightly down. Beating the fee is the whole game.
4.6M

Kalshi trades checked. People who take the posted price lose money in every category, from crypto to sports.

−$11.35

The bot's whole season, every rule and the forex quoter counted.

−$226.56

My own hand trades over 148 bets. The bot is the better trader of the two of us.

Data: Kalshi public trades and the bot's own ledger, season through Oct 17, 2026. Losses are measured at the real prices paid, fees included.

Sealed tests

Locked in before the exam

Each test is written down and sealed before it runs, with the rule, the dates and the pass mark fixed. That way a result can't be picked after the fact.

The questionExamStatusSeal
Can the bot's own odds beat the posted price on 17 quieter Kalshi markets?
fv-paper@1
Our call before the test: expect FAIL
Oct 22 – Oct 28 Sealed, exam starts Oct 22 1249f03567c099c6…
Do the bot's stock signals beat simply holding the S&P 500?
stock-signals@1
Before sealing it trailed the S&P 500 by 4.2% a month.
Nov 21 – Mar 31 Sealed, exam starts Nov 21 8ac2556298420daf…
When Kalshi and Polymarket disagree on the same window, is the gap tradable?
gap-paper@2 · kalshi-paper@1
The edge came from Polymarket's REST API serving books a median 11 seconds old. Re-priced on the live websocket, +5.5¢ a trade became −0.4¢.
to Oct 24 (voided Oct 20) Voided — stale data feed, caught before any money 5fb2120b6d223613…
Does the gauge that measures that gap read it correctly, live?
poly-watch@2
It read the same stale feed, so it was voided with the gap tests. Ten stale-source tests were voided in all.
from Oct 21 (voided Oct 20) Voided — stale data feed, caught before any money cf079a023478e103…
Killed edges
  • Crypto Lead market making and reward farmingdead (October exam, z 0.4)
  • Reward farming on slow marketsdead
  • The bot's own fair value vs the order bookthe book wins

Every seal is a fingerprint of the test file, taken before the exam. It gets published next to the result, so anyone can check nothing was changed.

Coming next

The 15-minute desk

One screen for the 15-minute markets. It shows what a seat really costs before you buy it, using the same math the bot runs. The seat-cost calculator and live board are free right now.

Bitcoin up in this 15 minutes? Example
Settles on60-second average at 11:00:00 ET
Right now+0.03% above the start
A YES seat costs54¢ + 2¢ fee
You need to win56% of the time
The bot's odds52%
SKIPThe seat costs more than the bot thinks it's worth.
The live settlement numberThe exact average the market settles on, and the move still needed.
The true cost of a seatPrice plus fee, and the win rate you need to break even.
The bot's odds beside the bookIts calibrated chance next to the market's price, with its track record attached.

Get on the waitlist

One email when the desk opens. Nothing else.

We store your email privately and never sell it. Privacy

Build a bot

Want your own trading bot built?

Live order-book feeds, settlement math, safety breakers and honest testing for Kalshi and Polymarket. The same stack that runs this bot, built around your idea.

Software only. We never hold your keys, and we never trade your account.

Tell us what to build

Every request is read by a person before anyone replies. Privacy