See the odds before you throw.
Most people lose money on Polymarket and Kalshi. We measure exactly how: what the fees really cost you, which trading bots actually work, how a market settles, and what we lost finding out. Free, no account, every number linked to its data.
- We don't trade. We did, lost, and published it
- No ad networks, no money from bot vendors
- Every number links to its data and code
True Cost: what the hoop really is
Prediction markets are zero-sum before fees and negative-sum after. Pick a venue, a price and how often you trade. The fee comes from the venue's own formula; everything else is arithmetic you can check.
To break even you need to be right —. On-chain studies put the share of wallets that lose money at about 84%.
Kalshi (schedule of 7 July 2026): 7% × contracts × price × (1 − price), rounded up, on orders that take liquidity; most sports and economics series also charge resting orders 1.75%. Polymarket US (schedule of 6 November 2026): Θ × price × (1 − price) per share on orders that take liquidity, where Θ is 9.5% for accounts under $10,000 of 30-day taker volume, 8.5% to $50,000, 7.5% to $250,000 and 6.95% above; new accounts pay 6.95% for their first seven days; resting orders pay nothing. "No edge" means your win rate equals the price, which is what the market is telling you it is. Sources: the venues' published fee schedules, linked from our repo.
Here's the deal, if you still want to play
Both venues run new-player bonuses, and both pay publishers when a new account signs up through their links. Below is exactly what happens, in order. If you're trading money you need this month, close this tab.
Kalshi
Trade $25, get $25 in credit- You open an account: name, address, date of birth, ID check. 18+, not in AZ, MN, NV or WA.
- You deposit (no minimum) and trade $25 in contracts within 90 days.
- Kalshi gives you $25 of trading credit. You cannot withdraw it; you have to trade it.
- Trading the $50 as 50¢ contracts costs about $1.75 in fees each way, before anyone wins or loses.
- We are not paid for this link yet. If Kalshi approves us as a partner, the rate appears here the day they tell us.
Polymarket US
Deposit $10, get a $25 bonus- You open an account with a photo ID and a selfie. 18+, not in Nevada. A promo code cannot be added later.
- You deposit at least $10 and place one real-money trade.
- You get a $25 bonus. It has to be traded once before it can be withdrawn.
- Fees on a 50¢ share: about 1.7¢ in your first week, then 2.4¢ (9.5% × price × (1 − price), the rate for everyone under $10,000 a month since 6 November 2026); resting orders are free.
- We are not paid for this link yet. If Polymarket approves us as a partner, the rate appears here the day they tell us.
Offer terms are the venues' and change without telling us; last checked 8 November 2026 against the venues' own promotion pages and the terms published by their partners. Problem gambling help: 1-800-GAMBLER.
Bot benchmark: is this bot legit?
About 115 trading bots are sold for Polymarket and Kalshi. Every review site we found is run by a vendor, and no one measures results. We run each bot's published rule on paper against recorded order books for 30 days and publish the number, the sample size and the data file. Vendors get an unedited reply under their row.
| Bot | What it claims | Price | Custody | Our 30-day paper result | Status |
|---|---|---|---|---|---|
| Our own Mr H (buy the spike) | Buy a 5-minute spike, trail a stop | Free, open source | Your machine | −1.86% a trade, 24% winners (1,172 signals) | Published |
| Our own Mr T (final seconds) | Buy the decided side in the last 5 s | Free, open source | Your machine | 0 fills possible in 368 windows | Published |
| PolymTradeBot | Momentum on BTC/ETH 5- and 15-minute markets | $499 once | Non-custodial | — | Queued |
| The "buy 45¢, sell 70¢" retail rule | Buy a side at 45¢ or less, sell it at 70¢, hold otherwise; BTC 15-minute. Circulated in tutorials and vendor marketing; not one vendor's product | Free to copy | — | — | Replaying on recorded books |
| Turbine reference bot (bot-builder platform) | The documented reference strategy: buy the side the live price favours vs. the strike (10 bps threshold, confidence-priced limits), Kalshi BTC 15-minute; not a product the vendor sells as a strategy | $9 to $199 a month plus compute | Hosted, your key | — | Replaying on recorded Kalshi books |
| AstroTick (open source) | Fee-aware favorite in the final two minutes, BTC 15-minute | Free | Your machine | — | Replaying on recorded books |
| Control: buy Up at the open | Not a product; the baseline every bot has to beat | — | — | — | Replaying on recorded books |
| Copy-trading bots (4 vendors) | Copy a named wallet | 0.25% to 1% a trade | Varies | — | Method in design |
| Cross-venue arbitrage scanners | 100+ arbs a day, 2 to 8% spreads | From $59 a month | Scanner only | — | Queued |
A dash means not measured yet, never "zero". Measured rows show net per dollar staked, the win rate, the number of trades, the worst losing run and the share of the window our books covered. How we test: the vendor's published rule, our recorded books, fees at the venue's real rate, fills only where the book showed size, and the coverage of our recording stated on the row. We make nothing from any bot on this list. Vendors: publish your rule and write to [email protected]; we never need your code or your keys.
Research: measured, with the code to check it
We publish what did not work as readily as what did.
-
8 November 2026
We ran the "buy at 45¢, sell at 70¢" rule on 689 recorded Bitcoin markets. It lost 9 cents on the dollar.
The most-quoted retail rule for 15-minute crypto markets, replayed on our recorded order books with the real fee on the real date: 48% winners, −$219 on $2,416 staked, and fees were three-quarters of the loss. The first row of the Bot Benchmark.
-
8 November 2026
We found free money in our own data. Then we tested it live.
192 "cheap winners" in the final seconds of Polymarket US crypto markets. With a live order book, the winning side was bid at 99¢ by the thousand and never offered. Where the mirage came from, with the raw books and the bug that fooled us twice.
-
8 November 2026
We bought the pump candle 1,172 times. Here is what 30 days of Coinbase data says.
24% winners, −1.86% a trade, and a median path that goes down the moment you buy. The signal is a coin flip; fees decide the sign. Plus the one green cell in the grid.
-
Next
The bot market: 115 products, zero independent tests
Who sells bots, how they charge, who got hacked, and why every review site is run by a vendor.
-
Learn (kept current)
Polymarket US fees, explained · Kalshi fees, explained · Kalshi vs. Polymarket US fees · How a 5-minute crypto market settles · Insider trading: how it works and how it gets caught · Is Kalshi safe? · Is Polymarket US safe? · Can I withdraw from Polymarket US? · When the market looks like the only way out · What the research says
The formulas from the venues' own schedules, worked at every price, the break-even math, and the settlement rule second by second. Each page carries the date we last checked the source.
-
Every market
13,000 market pages: the premise in plain words, what the market says, the fee at that price, and how much luck it needs
One page per market series on both venues, rebuilt hourly from the venues' public data. Real-world odds (sportsbook consensus, forecasts, the score) and the debate floor under each page are next.
Prediction News: the story, told by its sources, with the number attached
What is being reported about prediction markets, linked to the people reporting it, with one measurement of ours under each headline. We add the number; we don't add adjectives.
- 6 November 2026 · Polymarket US fee schedule
Polymarket US tiers its taker fee by volume: 9.5% for accounts under $10,000 a month, 6.95% only above $250,000 and for a new account's first week
Our number: A 37% fee increase for the ordinary account: $1.74 → $2.38 on 100 shares at 50¢; break-even 51.7% → 52.4%. More →
- 23 April 2026 · Mondaq, 5 May 2026 (CFTC v. Van Dyke; U.S. v. Van Dyke)
DOJ indictment and CFTC complaint over Polymarket trades on the Maduro operation using classified information
Our number: Alleged profit: more than $400,000, placed hours before the operation. Every dollar of it came from the accounts on the other side. More →
- 23 April 2026 · WBAY, 23 Apr 2026
Wisconsin's attorney general sues Kalshi, Polymarket, Robinhood, Coinbase and Crypto.com over sports event contracts
Our number: Sports is 59% of the market series on both venues in our index (7,760 of 13,182 pages). A ruling against sports contracts would touch most of what is traded. More →
Kalshi suspends three political candidates for five years for trading on their own races
Our number: Positions of about $50 to a few hundred dollars; fines of $540 to $6,229 plus disgorgement. The sums were small; the rule is not. More →
- 23 February 2026 · Front Office Sports
Kalshi removes all its affiliate badges on X after X bars paid affiliate accounts from posting gambling content
Our number: Category economics (iGB Affiliate, Nov 2025): about $100 per new trader to the publisher who sends them. Our referral links, if any, live only on this site, under the disclosure. More →
The Endorsement Ledger (who is paid to tell you to bet, with sources) is in review and will appear here.
Our ledger: we played too
From 2 to 8 November 2026 our bots traded real money at tiny size to test ideas. Every trade is below, read from the bots' public log files on GitHub. We stopped on 8 November when the last idea failed its live test. The record stays up because it is the point.
| Settled | Bot | Market | Side | In | Out | Net | Why it closed |
|---|---|---|---|---|---|---|---|
| Loading the record… | |||||||
Three ideas, three posts: buying the spike (lost), the final seconds (a mirage), decided games (shelved; it would only move money from slower players). Source files: mrh-live/live_trades.jsonl and mrt/trades.jsonl in the repo's data branches.
Data for people who check things
Our recorder stores one-second order books and index prints for every short crypto market on Polymarket US and Kalshi, with the settlement for each. Journalists and academics: write to us. Builders and agents: the API answers per call in USDC, no account, no key.
| Endpoint | What you get | Updates | Price per call |
|---|---|---|---|
GET /v1/preview | The last 3 settlement windows with their final books, 10 minutes delayed | Every 5 minutes | Free |
GET /v1/final-seconds/recent | The last 200 settlement windows with every book reading | Every 5 minutes | $0.002 |
GET /v1/final-seconds/stats | How efficient the book is in the final seconds, by coin, horizon and hour | Every 5 minutes | $0.002 |
| Dataset licence | Full recorder history with settlement records, for research or reporting | Daily | Ask |
The API goes live with the home server this week. We do not resell the venues' order books; we sell what we measure from them, cited to the venue. [email protected]
How we make money, and what we will not do
We make money from
- Referral payments from Kalshi and Polymarket US when you open an account through us, with the terms above the button. Current rate: not yet approved; posted here the day it is.
- Members ($8 a month, opening soon): the data files early, the methods, a monthly measured report. Everything on the site stays free.
- One sponsor slot, labeled. Never a bot vendor, never a venue we measure.
- Data licences and per-call API fees.
- Nothing else. If that changes, this list changes first.
We will not
- Trade. We stopped on 8 November 2026 and the record is above
- Hold your keys, your wallet or your money
- Take a cut of your trades, or money from any bot vendor
- Run ad networks, or sell your data or your email
- Show a venue's offer without the fees and the odds next to it
- Promise a return. Most prediction-market traders lose money; so did we
Plain answers
Are you telling me not to trade?
We're telling you what it costs and how often people win. Most lose. If you still want to, the deal section shows exactly what happens, including whether we get paid.
You plan to take referral money. Why trust you?
Because the fees and the odds sit above every button, the rate we're paid is published, and the block disappears before the warning ever would. Members and data buyers pay us to keep it that way.
Did your own bots make money?
No. The ledger is the real record: we lost a little, learned three things, published all three, and stopped.
Which venues?
Polymarket US and Kalshi, the two US-regulated exchanges. Not the international Polymarket book.
Can I check your numbers?
Yes. Every post links its data file and script, every ledger row comes from a public log line, every fee figure follows the venue's schedule, and the code is public on GitHub.
Is this financial advice?
No. It is a record of what rules say, what markets showed, and what fees cost. You make your own decisions.
Get the measurements
Research posts and new benchmark rows as they publish. Nothing else, and you can leave with one click.