How to Verify an AI Crypto Trading Bot's Track Record
A trading record is only as good as your ability to check it. Screenshots and dashboards prove nothing: the CFTC describes a fraud that faked customer balances with demo trading accounts. A backtest shows what rules would have done on old prices. A paper record on live prices is harder to fake but involves no real money. The strongest records combine every trade, losers included, with prices you can check against the exchange's own data, in a log where any later edit would show. Botsfolio publishes its trades that way: every filled trade, paper and real money, sealed every hour in a public log on GitHub. You can check it without an account.
In September 2026 the SEC charged two groups that drew investors in through WhatsApp and other platforms with talk of AI trading. In one of them, the SEC said, "there were no AI trading bots". In November the Alberta Securities Commission warned about scams that claim a trading bot can "produce high returns with little-to-no risk". Both kinds of pitch come with numbers: balances that climb, win rates, profit screenshots.
You can't judge a bot by its numbers until you know how the numbers were made. This page ranks the kinds of proof you will be shown, from weakest to strongest. Then it checks a real record, our own, step by step. We make Botsfolio, so weigh that. Every check works on any bot.
Seven kinds of proof, from weakest to strongest
| Kind of proof | What it shows | What it can't show | How to check it |
|---|---|---|---|
| 1. Screenshot or dashboard | A number someone chose to show you. | That any trade happened. The CFTC describes a fraud that "created fake customer accounts and balances using MetaTrader demo accounts". | Ask for the full list of trades behind the number. |
| 2. Backtest | What a set of rules would have done on past prices. | Real fills, or whether the rules were tuned until the past looked good. | Is it labelled as simulated? Are fees in? Was some history kept aside and never tuned on? |
| 3. Paper record on live prices | Trades placed as prices happened, from a dated start. | Real fills, slippage on thin coins or whether anyone stayed through a losing month. | Did it start before the results it shows? Do its prices match the exchange's candles? |
| 4. Profit card or leaderboard | One account's result, as the platform works it out. | What followers made, over what period, with what drop. | Look for the period, the worst drop and what followers made. |
| 5. Read-only exchange connection | An account's real fills, read from the exchange by a third party. | Whether this is the only account, or the best of many. | Who read it, from when and whether every account is included. |
| 6. On-chain history | Every trade an address made, readable from the address alone. | Who controls the address, or whether a bot traded it. | Look the address up yourself on the chain's explorer or API. |
| 7. Signed, time-stamped log of every trade | Who published each row, that no row changed since and when it existed. | On its own, that an exchange filled the order. | Check the signature and the timestamp. Then compare prices with the exchange's own data. |
A few rows need a word more.
Backtests. Rules can be tuned until the past looks good, which is why a backtest is a claim about the past, not a record. Regulators treat it that way too. The CFTC's rule for commodity trading advisers requires simulated results to carry a statement that "these results do not represent actual trading". AI tools now build backtests from a single sentence, and 3Commas, Coinrule and Bitsgap all offer one. AI made building a bot easy. It did not make the idea any better. More on why backtests flatter a strategy is in why trading bots fail live.
Leaderboards. A leaderboard shows the leader's account, not yours. In our study of every copy trading vault on Hyperliquid, 31% of followers were in profit after the profit share. The top 10 vaults by 30-day return lost a median 49% over the following year.
On-chain history. On Hyperliquid, the public API returns an account's fills from its address, with no key. Nobody can edit that history. What can mislead is the choice of address, when someone shows the one wallet that did well out of many.
Five questions that expose a weak record
Can I see every trade, or only a summary? A summary hides the losing trades and the stretches between wins. Bitsgap shows why it matters, to its credit. Its homepage shows an 11% average for its grid bot. Its own research found a median return of 0.51% for grid bots closed in the first half of 2026. An average and a typical result tell different stories.
When did the record start? A record that begins after the good results were known can be picked to look good. Look for a start date that came before the results. Then check that it keeps running.
Who measured it? The seller, the platform or the exchange? A number the seller typed proves less than one read from the exchange. When we read Cryptohopper's marketplace on November 13, 2026, its copy bot cards showed a profit figure with no period.
Is it real money, paper or a backtest? Each is useful and each is a different claim. A trustworthy page says which one you're looking at, next to every number. The gap between paper and real money is covered in paper trading vs real money.
Could it have been changed since? A spreadsheet or a web page can be edited quietly. A signed, time-stamped log shows any later edit. On-chain history can't be edited at all.
Once a record passes these questions, the next step is reading it well: every trade, a comparison with holding, the worst drop and the losing stretches. Those are the five checks for any trading record.
A worked example: checking Botsfolio's record
Here is the same test on our own record. Each step takes a few minutes. None needs an account.
- Open the log. Every filled trade, paper and real money, is in a public log on GitHub. A new seal is added every hour.
- Check who wrote it. Each seal is signed with Botsfolio's key. The matching public key is published on botsfolio.com and in a DNS record, which only the owner of the domain can change.
- Check that nothing changed. Each seal lists the SHA-256 of its files and of the seal before it. Editing or removing an old trade would break every seal after it. A correction goes in as a new row that names the row it fixes.
- Check when it existed. Each seal is tied to a Bitcoin block through OpenTimestamps, which nobody can backdate.
- Check the prices. Paper trades use live Coinbase prices. A buy at the daily close equals that day's close. A sell falls inside that minute's candle. Real-money rows carry the exchange, the fill price and the fee.
- Read the results built from those trades. Real money and paper trading, each next to holding the same cryptos.
The public log is on GitHub. The steps with the commands, including a script that compares the newest prices with Coinbase's public candles, are on how to verify our trade log. Here are the results built from those trades, as of the last daily close:
Results, and how each is measured
- Real money
- As of the Nov 17, 2026 close, real-money portfolios running on Botsfolio are down 5.13% since Oct 22, 2026, from real fills after all fees. Holding the same cryptos over the same days is down 24.82%. See real-money results
- Paper trading live markets
- As of the Nov 17, 2026 close, Botsfolio AI's own portfolios, trading live markets with paper money since Oct 8, 2026, are down 6.63% together after trading costs. Holding the same cryptos over the same days is down 26.47%. Real users' paper portfolios, run by the AI on cryptos they picked, are down 5.91%, while holding those cryptos is down 24.13%. Each trade happens as the market moves, not on past prices. No real money. See paper trading results
- Backtest
- The same rules from every monthly start since 2018, on past prices, next to holding the same cryptos. See the backtest
New trades are sealed every hour in a public log, signed by Botsfolio and anchored in Bitcoin. How to verify
Past results don't predict future ones.
What our log can't prove. A signed log shows who published each row, that none has changed and when it existed. On its own it can't prove that an exchange filled a real-money order. Real-money rows carry the exchange, the time, the fill price and the fee, so you can compare each one with that exchange's prices. They never carry a name, an account or a size, so you can't tie a row to a person. That protects the people trading with real money. It also means the rows are evidence, not a bank statement.
Where the money sits while a bot trades is a separate question from its record. That one is answered in can a trading bot withdraw your money.
If the bot trades stocks or forex
Most of this page applies to any market. Two things differ. Stock and forex accounts sit with regulated brokers, so a statement from the broker is strong evidence when you see the original. Forex traders often share results through tracking sites such as Myfxbook. Check whether a page's numbers were read from the broker account or typed in by the trader.
The CFTC's rule on simulated results applies to advisers in futures markets. Be wary of any simulated result shown without the statement that it doesn't represent actual trading.
- Every trade is listed, losers included. Not a summary, a chart or a screenshot.
- It started before its results. And it is still running.
- Someone other than the seller measured it. An exchange, a blockchain or a signed log.
- Real money, paper and backtest are labelled. Next to every number.
- Prices can be checked. Against the exchange's own data.
FAQ
Sources
Every page below was read on November 18, 2026. Pages change, so check the link before you rely on a line. If something here is wrong or out of date, tell us in the chat and it will be corrected.
- SEC press release 2026-95, September 29, 2026. Charges against two groups. In one, no AI trading bots existed
- Alberta Securities Commission, November 3, 2026. Investor alert on AI trading scams
- CFTC customer advisory on AI trading bots. Fake balances made with MetaTrader demo accounts
- 17 CFR 4.41 (Cornell Legal Information Institute). The statement required with simulated results
- Hyperliquid docs: Info endpoint. An account's fills from its address
- Bitsgap homepage. The 11% average for its grid bot
- Bitsgap: What Actually Worked in 2026?. The 0.51% median for grid bots closed in the first half of 2026
- Cryptohopper marketplace: copy bots. Profit figures with no period, read November 13, 2026
- Botsfolio trade log on GitHub. Every filled trade, sealed every hour
Watch Botsfolio AI trade live
Botsfolio AI paper trades on the live market, and every entry, stop and exit is public.
- BTC & ETH Cycle Trader+17.8% vs -33.1% buy & hold in 2026 · biggest drop 12.2% vs 46.5%
- BTC & ETH Trend Rider+1.0% vs -33.1% buy & hold in 2026 · biggest drop 8.7% vs 46.5%
- BTC & ETH Steady Trader+1.1% vs -33.1% buy & hold in 2026 · biggest drop 4.6% vs 46.5%
- Bitcoin Cycle Trader+0.2% vs -27.0% buy & hold in 2026 · biggest drop 7.6% vs 39.6%
- Crypto Cycle Trader+20.2% vs -14.2% buy & hold in 2026 · biggest drop 18.0% vs 45.5%
- Altcoin Trend Rider+21.6% vs -13.2% buy & hold in 2026 · biggest drop 20.1% vs 50.2%
- Memecoin Trend Rider-9.9% vs -52.8% buy & hold in 2026 · biggest drop 13.9% vs 65.0%
How to read these numbers: the buy and hold comparison, biggest drop, and how the portfolios are tested.
See everything Botsfolio AI trades live
Paper trading on real prices. Not investment advice.
Botsfolio is software. It is not investment advice and it is not tailored to you. Other companies' figures are as published on the dates given. Paper trading results are simulated. Trading crypto can lose money. Past performance does not predict future results.