1. When it acts
Every four hours, twenty seconds after a 4-hour candle closes on Binance: 00:00, 04:00, 08:00, 12:00, 16:00 and 20:00 UTC. It never trades in between. Stop-losses sit on the exchange itself, so they work while the bot sleeps.
2. What it trades
Ten large coins as USDT perpetual futures: BTC, ETH, BNB, XRP, SOL, DOGE, ADA, LINK, AVAX and DOT. It can hold each one long or short.
3. Sixteen questions per coin
For each coin the bot asks sixteen yes/no trend questions. Each answers +1 (up), −1 (down) or 0. They come in three families, so no single indicator decides:
- Trend, 5 votes: is a faster moving average above a slower one? Five pairs, from 1 day against 4 days up to 16 days against 64 days.
- Momentum, 5 votes: is the price higher than it was 7, 14, 30, 60 and 90 days ago?
- MACD, 6 votes: three MACD settings (12/26/9, 6/13/5 and 24/52/18 days), each asked two things: is it above zero, and is it above its own signal line?
The three family averages are averaged again into one score between −1 and +1. On the performance page you can see each held coin's sixteen votes as they stand now: green is up, red is down.
4. Bitcoin sets the tone
If Bitcoin is above its 50-day average, every short signal counts half. If it is below, every long signal counts half. The bot can still go against Bitcoin's trend, but it needs twice the conviction.
5. Four coins, one new one at a time
- A coin can be bought (or shorted) once its score reaches ±0.70. It is closed when the score falls below 0.40 or changes sign.
- At most four coins are held. If more qualify, the strongest score per unit of volatility goes first.
- At most one new position opens per four-hour cycle, so a copier's account builds up gradually instead of taking four trades at once.
Why four and not ten, or one? That took 144 backtests and a correlation study.
6. How big
- Calmer coins get bigger positions and wilder coins smaller ones, so each one carries a similar amount of risk.
- The whole portfolio aims to swing about 20% a year (annualised volatility). In quiet markets it holds more, in wild markets less.
- Hard limits: no coin above 25% of the account, all positions together never above 1.5× the account. Leverage is set to 5× cross margin, but in the backtest the account's real exposure averaged about 0.3× and peaked at about 1×.
- A position is only resized when its target moves more than 30% away from what is held. Smaller drifts are ignored, which saves fees. (We tested wider gaps and decided against them.)
7. Getting out
- Signal exit: the main exit. The trend fades below 0.40, the coin is sold at the next close.
- Stop-loss: a stop order on the exchange 25% from the entry price, on every position, at all times.
- Profit lock: once a position is up 100%, its stop moves to 25% below that price and never moves back down.
- No take-profits. We tested them; they cut off the few big trends that pay for everything else.
8. Safety limits
These only ever stop the bot from adding risk. Exits and stop-losses always go through.
- Kill switch: if the account falls 25% below its peak, every position target is halved until a human reviews it.
- Daily loss limit: after an 8% loss in one day (UTC), no new positions until midnight.
- Cooldown: a coin that hits its stop-loss can't be bought again for 24 hours.
- Funding filter: no new position on the side paying more than 0.1% funding per 8 hours.
- Liquidity cap: no position larger than 1% of the coin's daily trading volume.
- Order size: no single order needs more than 10,000 USDT of margin; bigger trades are split.
- Pause switch: one command stops all new entries while exits keep working.
9. How it places orders
New positions start as a maker order at the best price for 45 seconds (maker fees are lower), then cross the spread for whatever is left. Exits go straight to the market; waiting on exits cost more than it saved. Every fill is logged with its slippage.
10. What the backtest says, and what it doesn't
With all the rules above except the four newest safety limits (daily loss, cooldown, funding and liquidity, which aren't in the simulation), the backtest from January 2021 to September 2026 made 45.7% a year, with a Sharpe ratio of 1.67 and a worst drop of −18.5%, after 0.07% cost per trade. Adding Binance's real funding payments brings it to 41.8% a year (how we measured it).
* 2026 runs to 27 September. Hypothetical results; funding payments not included in the chart.
Read that chart carefully. 2021 was a once-in-a-cycle bull market, and the strategy did far better from 2021 to 2023 (Sharpe about 2.2) than from 2024 to 2026 (about 1.1). 2025 made about 7%. If you expect 2021 again, you will be disappointed. If you expect years like 2025, with occasional drops of 15–20%, you have the right picture.