Here is a thing most trading sites will never tell you: the boring version usually wins, and the clever additions usually make it worse. We believe that, but belief is cheap. So we did the work — we tried to beat our own bot, and we wrote down what happened whether it flattered us or not.
A quick word on the bar. The bot’s current strategy turns $10,000 into about $338,000 across 11 years of Bitcoin history in our simulator, with a worst-case fall of roughly 23% along the way. For any change to “win,” it had to clear a pre-registered gate set before we looked at results: meaningfully better risk-adjusted return, no deeper drawdown, and — the part that kills most ideas — it had to keep working on years of recent data we deliberately hid from the tuning.
Attempt 1: stop paying so much to trade
Trading isn’t free — fees and slippage quietly bleed a strategy every time it moves. We measured it: those costs drag the bot by almost 9% a year, and a cost-free version of the exact same strategy would have ended up 2.6× richer. That’s a huge number, so we built a filter to make the bot trade less — wait for bigger moves, ignore the small stuff, hold positions longer. Thirty-seven different versions of it.
Result: 0 of 37 passed. The version that genuinely cut trading by a third still ended up poorer — it saved on fees but missed the start of the moves that mattered. The lesson stung a little: that 9%-a-year cost isn’t waste, it’s the price of being responsive. You can’t skip the fee without skipping the trade.
Attempt 2: sit out the dead zones
A separate study (the candle-patterns one) had found a genuine signal: when Bitcoin’s volatility quietly dries up while price is below its recent average, the market’s usual drift tends to stall for weeks. So we told the bot to shrink its bets during those dead zones.
Result: 0 of 6 passed. The catch: the bot’s trend rules already keep it out of those zones almost entirely — the dead zone overlapped only about 4–7% of the time it was actually holding a position. Damping that sliver cost more than it saved. A real signal, measured correctly, that the strategy simply didn’t need.
Attempt 3: buy the bottom
This is the one that looked like a winner. Using an on-chain measure of how stretched the market is (how far price sits above what holders actually paid), we told the bot to buy when Bitcoin looked truly capitulated — blood-in-the-streets cheap. On paper it ended at $656,000 instead of $338,000 — nearly double — at the same worst-case drawdown.
Why we killed a result that doubled the money
Because it rested on about four events in eleven years. True capitulation is rare, and one of those four — the March 2020 COVID crash — actually lost money when the rule fired, buying a knife that kept falling. A strategy whose whole edge comes from three or four moments isn’t an edge; it’s a bet that happened to land. And when we loosened the “cheap enough” threshold even slightly, the worst-case drawdown blew out to 41%. We’d rather leave $300k of backtest profit on the table than ship a coin flip.Result: 0 of 4 passed. The most tempting failure of the day.
Attempt 4: fade the crowd
When everyone piles into the same leveraged bet, it often ends badly — so we measured crowding through funding rates (what leveraged traders pay to hold their positions) and told the bot to pull back when the crowd ran hot.
Result: 0 of 4 passed. Every version ended poorer, and the failure was the same shape as Attempt 1: the crowded, expensive-funding moments cluster inside the biggest bull runs. Pulling back from the heat meant pulling back from the rallies. The signal was real; trading on it was a mistake.
What four failures actually buy you
A day of work, four good ideas, zero improvements — and that is a genuinely useful outcome. We didn’t just fail to beat the bot; we learned why it’s hard to beat. Its costs are the price of its speed. Its dead zones are already avoided. Its blind spot (calling exact bottoms) can’t be fixed with the few examples history offers. The boring strategy survived four honest attempts on its life, and we trust it a little more for it.
The honest conclusion
Most “trading secrets” are someone’s best idea that never got tested against the years it would have failed. We test ours, we hide recent data from the tuning so we can’t fool ourselves, and we publish the failures — especially the one that doubled the money. A result you can’t trust is worth less than a clean “no.”