Should you set up your futures grid bot the way the video says? On 18 October 2026, Furkan Yildirim published a rulebook on his YouTube channel CoinCheck TV, derived with Claude from more than 700,000 simulated futures grids: 5-minute candles for Bitcoin, Ethereum, Solana and XRP from January 2024, for gold from March 2025, a new bot every week, 90 days each. The headline numbers: a long grid at 5× leverage on the "balanced" setting returns, averaged over these five markets, +71 % on margin in 90 days when price rises and costs −20 % when it falls. He hands out the matching settings tool to users who open or link an OKX account through his partner link.
That sounds like a bet with a built-in parachute: three parts gain against one part loss. Anyone who believes it sets the bot up exactly like that and expects to lose a fifth in the bad case.
The rules measure the range in daily swings
The grid range is measured in a unit U: ten times the average daily range of the last 14 days (ATR), relative to price. Two settings:
- careful, 3×: 1.5 U below the entry, 3 U above
- balanced, 5×: 0.5 U below the entry, 3 U above, no margin reserve
Plus: grid spacing never below 0.3 %, geometric rather than arithmetic.
Rebuilt on BTC, ETH and SOL: one start every Monday from January 2024 to July 2026, 90 days each, 134 starts per coin. Long grid with an initial position, the way Binance and KuCoin run futures grids, 0.05 % fee per order, funding included. The rules for the evaluation and the verdict were fixed before the first run started.
When price rises, the video is right
| 5× balanced, rising windows | Mean | 90 % range |
|---|---|---|
| BTC (59 windows) | +63.7 % | +45.4 … +73.8 % |
| ETH (56 windows) | +74.0 % | +53.6 … +89.9 % |
| SOL (59 windows) | +73.3 % | +54.4 … +80.6 % |
On all three coins the +71 % sits inside the range. The number is not made up.
It is not grid income, though. A grid without leverage on the same range earns a median +2.4 % (BTC), +0.5 % (ETH) and +4.1 % (SOL). The +71 % is the price gain of the initial position, levered five times. The grid does not collect the 71 %. It rides it.
When price falls, it is not
| 5× balanced, falling windows | Mean | 90 % range | liquidated |
|---|---|---|---|
| BTC (57 windows) | −59.4 % | −74.6 … −40.4 % | 23 of 57 |
| ETH (68 windows) | −85.8 % | −92.4 … −76.8 % | 53 of 68 |
| SOL (63 windows) | −77.7 % | −87.2 … −65.9 % | 42 of 63 |
Instead of −20 %, the same setting loses 59 to 86 % of the margin on average. On no coin does −20 % come anywhere near the range.
The reason is in the rule itself. The range reaches only 0.5 U below the entry. Once price falls below it, the bot holds its full position at five times leverage and the grid stops buying. On ETH that ends in liquidation in 53 of 68 falling windows. A parachute that opens below the ground is not a parachute.
The careful setting drops out when markets get rough
For the careful setting at 3× leverage, the video gives +35 % and −8 %, with the margin reserve switched on and no liquidation. In the rebuild it is indeed not liquidated in a single falling window, because its range reaches 1.5 U down.
That depth is the catch. When the daily range exceeds 6.7 % of price, the lower bound drops below zero, and the rule cannot set the bot up at all. That happened in 52 of 134 starts on SOL, 27 on ETH and 5 on BTC. The careful rule fails exactly in the phases that call for care.
Across all windows it turns into a coin toss
Nobody knows at the start whether the next 90 days will rise. Across all windows:
| 5× balanced, all windows | Mean | Median | liquidated |
|---|---|---|---|
| BTC | +2.9 % | +6.6 % | 24 of 134 (18 %) |
| ETH | −13.7 % | −20.1 % | 56 of 134 (42 %) |
| SOL | −4.1 % | +8.8 % | 43 of 134 (32 %) |
On top comes funding: at 5×, a median 3.3 % (BTC) to 8.3 % (ETH, SOL) of the margin goes to the other side every 90 days.
The video's arithmetic, three parts gain against one part loss, turns into a coin toss with a bad payout. Splitting by rising and falling is hindsight. You have to act before.
Geometric is not simply better
The video recommends geometric spacing for long grids, because it loses less in downturns. In the rebuild's falling windows that was one percentage point on BTC and none on ETH and SOL. Compared pair by pair, start by start, there is no detectable difference across all windows. Split up, a pattern shows:
- rising windows: arithmetic better, by a median 5 to 8 percentage points
- sideways windows: geometric better, by 2 to 5 percentage points
- falling windows: equal, because both variants are liquidated together
Which spacing wins depends on the price direction, which you do not know at the start.
An average with gold is not your coin's number
"The video ran 700,000 grids on OKX, you ran a few hundred on Binance prices."
True, and the technical differences are named: 15-minute instead of 5-minute candles, Binance instead of OKX prices and funding, and the video's exact grid count is unknown. None of these explains a gap from −20 % to −60 to −86 %. The high and low of a 15-minute candle contain every 5-minute high and low. Finer candles would not make liquidations rarer.
One difference weighs more. According to the video, the +71 % and −20 % are an average over Bitcoin, Ethereum, Solana, XRP and gold. Gold moves far less than crypto; a 30 % range is "huge for gold and very tight for Solana", in Yildirim's own words. Whether gold and XRP pull the average towards −20 % has not been recomputed here. The video also does not say when a window counts as "wrong". Running a bot on Bitcoin, Ethereum or Solana, you need your coin's number, not an average with gold.
Sample size is a separate point: the 134 starts per coin overlap, because every window runs 90 days and a new one starts each week. Roughly ten per coin are independent. The ranges in the tables are computed with that in mind.
The range below your entry decides
If you run a long grid at 5× leverage, do not budget the downside at −20 %. In the tested 2024 to 2026 period, the "balanced" range lost between 59 and 86 %, with liquidation in 40 to 78 % of falling windows.
The deciding setting is not leverage alone but how far the range reaches below the entry. In an earlier measurement a 5× grid was rarely liquidated, because the entry sat mid-range and the bot started without a position (Why your 5× grid bot rarely gets liquidated). With an initial position and a range that barely reaches below the entry, it is the reverse.
Run your own case in the grid backtester: leverage, start rule "initial position", and a range you fix beforehand, not afterwards. What a grid protects against in the first place is in Your grid bot is not a hedge.
The backtester cannot model the video's short grids and 10× leverage. Those claims remain untested.
FAQ
What exactly is U? Ten times the average daily range (ATR) of the 14 completed days before the start, divided by the entry price. The median U was 35 % (BTC), 50 % (ETH) and 62 % (SOL) of price.
What does "initial position" mean? Futures grids on Binance and KuCoin place one buy for every grid level above the entry at the start. The bot holds a position from the first moment and carries funding and liquidation risk from then on.
When does a window count as rising or falling? When the close on day 90 is more than 5 % above or below the entry. Windows in between are sideways and do not enter either test.
Why does the mean sometimes differ sharply from the median? Liquidations set single windows to −100 %. For ETH in falling windows, even the median is −100 %.
Which fee and which funding? 0.05 % per order. Funding for BTC from the recorded daily rate of three exchanges, for ETH and SOL a flat 0.05 % per day.
Is the result reproducible? Yes. A second run produced bit-identical results. The study is listed in the platform's falsification register.
Not investment advice, not a recommendation, not a forecast — historical patterns are no guarantee.
Study the Past — Improve your Future 🥋