Seven trades can look like proof. On a chart, with the entries marked and the targets hit, they feel like a strategy that works.
That was the starting point for Episode 1 of Creating Bots, our new weekly series. Each week we take a famous public strategy, turn it into rules with Pulse and test it honestly. First up: Data Trader's 4-hour range strategy, from his video The BEST 5 Minute Scalping Strategy Ever (Simple and Proven), which has about 1.9 million views.
The claim we set out to test
Here is his result, in his own words:
"Out of seven trades we took, the strategy gave us five wins and two losses with a total gain of 8R. That's about a 72% win rate on the crypto market."
Credit where it's due. Data Trader published rules clear enough to test, and he said himself that seven trades needed a bigger sample. That's exactly what we gave it.
The 4-hour range strategy in 60 seconds
His idea is simple, which is part of the appeal:
Mark the high and low of the first 4-hour candle of the New York day. That's your range.
On the 5-minute chart, wait for a candle to close outside the range. A wick doesn't count.
Wait for a later candle to close back inside, on the same day.
Fade the move. Broke the high and came back in? Short. Broke the low and came back in? Long.
Put the stop at the extreme of the breakout, with a target of two times the stop (2R).
Easy to follow by hand. Turning it into a bot is where the real questions start.
Three gaps a bot can't guess
We described the strategy to Pulse in plain English, the way you'd explain it to another trader. Pulse wrote it back as rules you can see and edit, and three of them needed a decision the video never quite made.
Gap 1: which 4-hour candle?
Crypto trades 24 hours a day, seven days a week, so there's no opening bell. "The first 4-hour candle in New York time" also moves against UTC when the US clocks change. Our rule: the range is 00:00 to 04:00 New York, following the clock changes, and no breakout counts until that block closes.
Gap 2: where exactly does the stop go?
His checklist says the extreme of the breakout. In some of his examples, though, the stop sits at the range high, or at "the nearest key level" when the breakout is big. A key level isn't a rule, because two traders would put it in two different places. So our base rule is the breakout extreme. A second version, Variant B, skips any trade whose stop is wider than three times the 5-minute ATR(14). That one gets its own test in a later episode.
Gap 3: what happens at the end of the day?
His levels are valid until the New York day ends, yet the video doesn't say what to do with a trade that's still open. Our rule: close it at 00:00 New York, with no new entries after that.
If you trade this by hand, you're making these three choices anyway. You just haven't written them down.
How we tested it
With the rules complete, we ran them in Strategy Lab:
Markets: BTC/USDT and ETH/USDT on 5-minute candles
Period: 1 October 2023 to 30 September 2026, three full years
Costs: fees and slippage on every entry and exit
Account: one shared account, risking 1% per entry, with a maximum of three open trades
Evidence Lab then ran each year separately. Within every year, it builds on the first 70% of the history and keeps the last 30% back as an unseen hold-out, data the rules are never judged on until the reveal.
Results
Full replay: 4,634 trades instead of 7
Across the three years, Bitcoin took 2,329 trades and averaged −0.30R per trade, with a profit factor of 0.57. Ethereum took 2,305 trades and averaged −0.22R per trade. Put together, that's 4,634 trades against the seven in the video.
A profit factor below 1 means the losing trades cost more than the winning trades made. At 0.57, every 1 USDT lost was matched by about 0.57 USDT won.
Each year, before and after costs
Year (1 Oct to 30 Sep) | Build-period trades | Before costs (USDT) | Costs (USDT) | After costs (USDT) | Profit factor | Max drawdown | Unseen hold-out (USDT) | Hold-out profit factor |
|---|---|---|---|---|---|---|---|---|
2023–24 | 1,044 | −8,678.86 | 58,784.80 | −67,463.34 | 0.58 | 67.8% | −44,721.56 | 0.58 |
2024–25 | 1,125 | −3,341.29 | 66,788.68 | −70,129.94 | 0.64 | 70.5% | −45,937.14 | 0.56 |
2025–26 | 1,013 | +8,425.73 | 71,317.30 | −62,891.87 | 0.67 | 63.1% | −35,273.69 | 0.55 |
Shared BTC/USDT and ETH/USDT account, 5-minute candles, 1% risk per entry, maximum three open trades. Build period is the first 70% of each year; the hold-out is the final 30%.
Costs are huge on a 5-minute strategy, because it trades so often. Still, look at the before-costs column. Only 2025–26 was ahead before fees and slippage, and the other two years lost money even with costs taken out. So costs alone don't explain the losses. The entries themselves weren't finding an edge.
Unseen data
Every hold-out lost money, with profit factors between 0.55 and 0.58. Bitcoin on its own lost 34.2% in the 2023-24 hold-out. Unseen data is the closest a back test gets to the future, and the rules didn't hold up there either.
Against random entries
Here's the question that matters most. Does fading a range break do any better than entering at random?
Evidence Lab built 1,000 random-entry versions with the same stop, target and exits. Our real entries beat 53.4% of them. That's about the same as a coin flip.
The scorecard
Luck test: Fail
Unseen data: Fail
Fragility: Fail, as 0% of stress runs stayed profitable
Bad-luck stress: Caution, with a worst-case drawdown of 63.9%
Plan status came back as Needs work, and the "passes your rules" check said No. Evidence Lab graded it F, 12 out of 100, in every year.
Seven trades vs 4,634: what this does and doesn't mean
None of this says Data Trader did anything wrong. A video shows examples, and examples are chosen after you know how they ended. Seven trades is a demonstration. Over 4,600 is a test.
It also doesn't prove every range strategy fails. Our version used our choices for the candle, the stop and the day end. Change any of those and you have a different strategy, which deserves its own test. That's the whole point of writing the rules down.
A back test is the past. A simulation is practice. Neither predicts future performance.
What happens next: a simulated bot
Our version failed the tests that matter, so it isn't going anywhere near real money. Practice still shows how a strategy behaves day to day, so on 23 October 2026 we started a simulated bot on the same rules. It runs on real market data, with fees and slippage included, and no real funds at risk.
A few days can't tell you anything, so we'll bring its results in next week's episode.
Test your own version with Pulse
Got a range strategy, or any idea you trade by hand? Describe it to Pulse in plain English and it will turn it into rules you can see and edit. Back test them in Strategy Lab, then let Evidence Lab check for luck, unseen data and fragility.
The Free plan includes Pulse, Strategy Lab, backtesting and one simulated bot. From the Pro plan, at $29 a month, live bots run on Bybit, Binance, OKX and Bitget, with Hyperliquid coming soon. Live trading involves risk and you can lose money.
FAQ
Does Data Trader's 4-hour range strategy work on crypto?
In our version, no. Over three years of BTC/USDT and ETH/USDT with fees and slippage, it lost money in every year and every hold-out, and Evidence Lab graded it F, 12 out of 100. Other rule choices could behave differently, so test your own version before you trust it.
Why test 4,634 trades when the video showed seven?
With only seven trades, a lucky run looks just like a real edge. Thousands of trades across different market conditions, plus unseen data and random entries, tell you far more.
Did fees cause the losses?
Not on their own. Costs were large because a 5-minute strategy trades so often, yet two of the three years were already losing before costs. Only 2025–26 was ahead before fees and slippage.
Can I back test this strategy myself for free?
Yes. On the Free plan you can describe it to Pulse, edit the rules, back test them in Strategy Lab and run one simulated bot, with no real funds involved.
Educational content only. Not financial advice. Trading involves risk.
