FoundLuxWhat it is sorted by
Not by the highest profit — that would wash lucky hits to the top. Rated are hit rate, profit factor, the number of trades, stability over time and the ratio of target to stop. That is divided by the deepest drawdown.
Test
Backtest & optimizer
Run each of the eight bots against real historical prices — with fees, with slippage, over the period of your choice. And then the more important part of this page: what such a result is worth and what it is not.
Trades on real candlesDemonstration
The percentages at the markers are calculated from the candles shown, not entered — that is why there is a loss among them too.
In the big chart
This is what a backtest looks like — for every bot
After testing, “Show trades in the big chart” puts every entry and exit on the candles, with the result. That way you see whether the bot does what you expect of it — and where it gets it wrong. Shown for each bot is the setting of its bot tour: BTC, twelve months, costs as live.
Basket study
Where the bots hold up — and where not
Every bot with its presets on all 135 coins, in three baskets: the twelve large ones (Majors), the twelve memes with the most turnover and all 112 others. Tuned on 2020 to mid-2024, confirmed on July 2024 to today — a period no setting had seen before. Costs as live, whole contracts.
| Bot · preset | Majors (12) | Memes (12) | others (112) | Assessment |
|---|---|---|---|---|
| Trend Guard · Standard 1d | 12 / 12 | 6 / 12 | 49 / 111 | holds up most broadly — Majors 12 out of 12 in both periods |
| Donchian · Calm 6h | 10 / 12 | 6 / 12 | 29 / 111 | built for the large coins |
| Confluence · Balanced 4h | 8 / 12 | 5 / 12 | 67 / 111 | broad on the other coins |
| Confluence · Safe 4h (score ≥ 4) | 2 / 7 | 4 / 7 | 41 / 54 | trades rarely, is then usually right |
| Grid · Wide ± 25 % | 8 / 12 | 4 / 12 | 27 / 109 | strong in rising markets (learning period 12/12 · 9/11 · 67/102), weak in falling ones |
| Scout · Quorum · DCA | 0–2 / 12 | 0–4 / 12 | 0–19 / 111 | hold up with no preset — the setup says so in yellow |
Figures of the test period (07/2024 – 09/2026): coins in profit out of coins with trades. With “Safe” many coins do not trade at all — hence the smaller numbers. The calculation is open: the same interface as the bot setup, $100 margin × 3 (Grid $500 × 1, DCA $100 entry, Trend Guard $100 × 1).
Long timeframes hold up
4h, 6h, 12h, 1d: few, good trades. On 15m and 1h every round costs 0.28 % — hundreds of rounds eat every edge. Confluence on 15m was not in profit on any of the 135 coins.
Memes are the hardest basket
They often run on single huge jumps. No bot holds 9 out of 12 there in both periods — few, large trades do best.
Check for yourself
Every bot setup has the three baskets: Majors, Meme and “All others” — the same setting across 112 coins, with the result per coin and the number of how many hold.
What you set
| Input | Meaning |
|---|---|
| Market | One of 140 — the bot is tested on exactly this one |
| Period | From–to, freely selectable. The longer, the more market phases |
| Timeframe | On which level the bot decides, 1m to 1d |
| Stake | Starting capital of the run |
| Leverage | Goes into profit, loss and liquidation |
| Fees | Deducted per trade — without them every result is embellished |
What you get back
Result after fees$ and %
Hit rateShare of winning trades
Profit factorSum of profits ÷ sum of losses
Deepest drawdowndeepest trough of the curve
Number of tradesStatistical significance
Long / short separatelydoes it earn in both directions?
Individual tradesexpandable list
The deepest drawdown is the number that counts. A result of +40 % is worthless if half the account was gone in between — in reality you would have switched it off before.
Automatic search
The Top 50 optimizer
Instead of trying out settings by hand, the optimizer calculates them systematically — for the Confluence bot that is 142,560 combinations, each across up to eleven timeframes. At the end the best fifty are shown as a table, from which you take one.
Split period
The period is cut in half: the settings are searched in the first half and tested in the second. What was only good in the first half fails. That is the most effective single protection against fooling yourself.
Apparent variety removed
Many combinations lead to the same trades in practice — for example because a distant target is never reached. Such twins are merged, otherwise fifty rows would fill up with the same setting.
A run takes time. Depending on market, period and timeframe, between a few minutes and almost an hour. It keeps running on the server even if you close the tab — you see the progress and can cancel.
The actual point
Why a good backtest is no proof
For every market and every period there are settings that look excellent in hindsight. Finding them is no art, just computing time. The only question is always whether they still hold tomorrow — and no backtest in the world answers that.
Overfitting
Anyone who tries 142,560 combinations inevitably finds some that look good — even if it is pure chance. With so many attempts that is not a possibility but statistics. The split period and the stability score slow that down, but they do not abolish it.
The market changes its character
A setting that mastered two years of sideways movement can give everything back in the first trend phase. No historical period contains the market phase that comes next.
Execution is kinder in the test than in reality
Fees and a surcharge for slippage are included. Not reproduced are partial fills, price spikes in thin moments and the fact that precisely in hectic phases the order is filled worse than assumed. A backtest is therefore systematically a little too optimistic — not randomly, but in one direction.
Few trades prove nothing
A hit rate of 80 % from ten trades is an anecdote. That is why the number goes into the rating — but check for yourself how many trades are behind a row before you take it over.
One market is one market
A setting that only works on Bitcoin and nowhere else has probably memorised Bitcoin's past. Test it on three or four other markets. If it roughly holds there, that is a much better sign than any number from the first run.
How to use it: backtest to weed out, not to pick. It reliably tells you what does not work. What is left you let run in demo mode — for several weeks, and long enough to have seen a bad phase. Only that is proof. To the eight bots →
Run the numbers yourself
Create an account, put a bot together, pick a period, let it run. No real money, no payment details.