Oct 24, 2026 · 5 min read
Ten paper trades prove almost nothing: a coin flip wins 6 or more of 10 about 38% of the time. Here is what to write down, how many trades to wait for and what a simulator hides.
TradingWizard AI5 min read
To test an AI trading setup before risking money, write the rules down first, run it on live prices with simulated money, and judge it on at least 50 to 100 decisions. Ten paper trades prove almost nothing: a coin flip gives you 6 or more winners out of 10 about 38% of the time.
That is the part most guides skip. They tell you to "paper trade first" and stop there. This page shows what to write down, how many trades to wait for, and which costs a simulator hides.
Hypothetical example. A bot with no edge at all wins 50% of its trades. After 10 simulated trades, here is how often it looks good by luck alone:
| Result after 10 trades | Chance with a 50% win rate |
|---|---|
| 6 or more wins | 38% (386 in 1,024) |
| 7 or more wins | 17% |
| 8 or more wins | 5.5% |
So "6 wins out of 10" is what random trading produces more than a third of the time. At 100 trades, a no-edge bot reaches 59 or more wins only 4.4% of the time. That is why a sample of 50 to 100 decisions, across more than one kind of market, is the minimum before a result means anything. It is still not proof, only a result that luck alone rarely produces.
Write these on one page before the bot runs, so the result cannot change the rules:
Change only one of these between test rounds. If you change three things after a loss, you cannot tell which one helped.
Hypothetical example. A setup risks $100 to make $150 and wins 40% of the time. Per trade that is 0.40 × $150 − 0.60 × $100 = $0. It breaks even. Now add $5 of fees and slippage per trade, which a simple simulator may not charge. The result is −$5 per trade, or −$250 over 50 trades, from a setup that looked flat.
Simulators differ in what they model. Alpaca's paper-trading documentation says its simulation does not account for market impact, latency slippage, queue position or regulatory fees. Check what your platform skips before you trust its fills.
For each decision, write down the market, the time, the entry, the stop, the target, and why the bot acted or waited. A skipped trade is data. If the bot never waits, it is probably not filtering anything.
After 50 to 100 decisions, compare the record with the page you wrote at the start:
A good simulated result is a reason to keep testing, and a live switch should wait for its own decision. FINRA warns that some automated-trading promotions make misleading claims about AI and returns, so be wary of any service that quotes its results as certain. If you later go live somewhere, make it a deliberate step with an amount you can afford to lose.
TradingWizard bots run in demo. Demo trades use simulated money. No real orders are sent. Each bot scan shows the setup state, the entry, stop and target when a setup exists, and the reason it traded or waited, so the review above can be done from the bot's own record. A scan can say Wait, and that counts as a decision.
Set up a bot on BTC in Wiz and read why it traded or waited.
TradingWizard is not a financial adviser. Nothing here is a recommendation to buy or sell.
Oct 24, 2026 · 5 min read
Oct 11, 2026 · 4 min read
Aug 18, 2026 · 8 min read
Jul 10, 2026 · 3 min read