Transparency
We don’t sell a smooth curve. We show you reality.
Among trading robots, the prettiest chart is usually the most dangerous. A curve that climbs month after month without a pause is, more often than not, one of three things:
Overfitting
The robot was tuned for the past, not the future.
Hidden risk
Martingale, grid, or no stop loss.
Unrealistic costs
A backtest without real spreads and swaps.
LFP has losing months and time under water, and we put both in plain view. Here is why.
No Hollow Promises. Just Proven Math.
Ask before buying any robot
Five signs of an overfitted robot
| Sign | What it hides |
|---|---|
| Almost no losing months | Losses have usually been postponed — losing trades left open, or size added (martingale or grid) — until the day they burn the account at once. |
| A profit factor of 3, 5 or more | In real markets with real costs, a figure like that almost always comes from over-tuning on the past or ignoring the spread. |
| One market and one short period | The robot may only have worked in that one market regime. |
| No test on data the design never saw | Without an independent test period, luck cannot be told apart from skill. |
| Costs are not stated | The same strategy can turn from profit to loss with real spreads; we saw exactly that on our own previous portfolio. |
The LFP method
What LFP does differently
- 1
Tested on real costs, not dream costs
- Every historical candle is simulated with the broker’s real spread at that hour of the day, and overnight swaps are charged.
- A real example: a portfolio that showed +243% on a demo account’s data lost 40% with the live account’s real spreads. We dropped it and rebuilt the robot from scratch on real costs only.
- 2
Tested on data the design never saw
- Settings were chosen on 2020–2023 data only. Data from 2024 onward was for testing alone.
- The pass rule was written before the result was seen. The one time we found our choice had been unintentionally influenced by the test period, we measured it and corrected it.
- 3
Most ideas were rejected — and that is good news
Test Tried Accepted Strategies on new symbols 31 3 Calendar patterns (tested statistically against chance) 11 0 Pairs trading and mean reversion 215 4 symbols in one strategy - A robot whose every idea “worked” has probably tested none of them properly.
- 4
6 independent strategies, not one big bet
- Three different logics — trend following, range breakout and mean reversion — across 8 markets (gold and 7 currency pairs).
- The strategies’ monthly correlation with one another is low, and the mean-reversion part is even negatively correlated with the rest.
- Each strategy is weighted so that none carries a disproportionate share of the risk.
- 5
No martingale, no grid, a stop loss on every trade
- No trade is enlarged because of a previous loss.
- Losing positions are never averaged down.
- Every trade has a defined stop loss, and its risk is known before it opens.
- The expert advisor also has a controlled recovery system; unlike martingale, it enters only on the strategy’s own signal, and its number of steps, per-trade risk and total basket risk are all capped. It is switched off in the current trading styles and in every published backtest.
- 6
The expert advisor is exactly what was tested
- The expert advisor’s trades in MetaTrader were compared one by one with the research simulation: 40 out of 40 trades matched.
- Each trading style’s settings are generated automatically from the same test data, leaving no room for hand-transfer errors.
- 7
A built-in watchdog: the robot knows when not to trade
- High spread: if a market’s spread exceeds 2× the level it was tested on, that market stands down temporarily.
- Broker changes: if the broker changes a contract’s specification, that market stands down and is reported.
- Fading performance: if a market is no longer profitable over its last 100 trades, it is set aside.
- A real example: our broker’s Nikkei index spread rose about 100-fold from June 2025. The watchdog stood it down, and we removed that strategy from the portfolio rather than hiding it.
- The robot never “re-optimises” itself on recent months. Re-tuning on one bad quarter is the fastest way to break a healthy system.
- 8
Account guards that match your goal
Style For whom Guard Real drawdown in backtest PropFirm Prop-firm accounts Daily and overall loss limits (default 4% and 10%). Before every trade it checks that open risk stays within the limit. 4.2% Brave Balanced growth; the default style A brake that closes everything at a 30% drawdown and resumes under 20% 20.6% Reckless High risk No brake and no loss limit; the permanent stop at 40% is optional and off by default 40.2% Mad Only for someone who knowingly accepts the risk of losing the whole account No guard 51.4% The full result of every measured style, from return to month by month, is in the LFP Console. Console
The numbers, without make-up
Backtest of the Brave style, with real costs
Backtest from 2020/01 onward
- Profitable calendar years
- 7 of 7
- Profitable 12-month periods
- 97%
- Profitable months
- 72% (58 of 81)
- Average up / down month
- +9.6% / −4.3%
- Worst month
- −10.1%
- Longest run of losing months
- 4 months
- Losing quarters
- 6 of 27
- Deepest fall from a high
- 20.6%
- Longest time below a previous high
- 435 days
- Profit factor
- 1.48
- Winning trades
- 53%
- Average trades per month
- 18
These are backtest results, not a live record. Half of the period was design data. Past performance, simulated performance above all, is no guarantee of the future.
Hard questions
Before you ask
Why does the robot stay below its previous high for up to 435 days?
Because every real system has such periods. The only question is whether they are shown.
- Compared with the market: the S&P 500, the world’s best-known investment, stayed below its high for about two years after January 2022, and for more than five years after the 2008 crisis.
- Why it is unavoidable: how long a system stays under water depends on its ratio of return to volatility. Under-water spells of a few days need a ratio seen only in high-frequency trading with million-dollar infrastructure.
- “Under water” does not mean “at a loss”: it means the account has not yet returned to its previous high. In LFP’s backtest 97% of 12-month periods were profitable; whichever month you started in, in 97% of cases you were in profit a year later.
- A robot that “never” goes under water has usually hidden its losses, in open positions without stops or in a grid.
Why are 28% of months negative?
- Simple statistics: with about 18 trades a month and a 53% win rate, some months are naturally negative. One month is a small sample; over each calendar year, every year was profitable.
- The losing months are small: on average about half a winning month (−4.3% against +9.6%).
- The alternative is worse: a system with no losing months has usually postponed losses, not removed them. We would rather show a small, controlled loss each month than give it all back in one day.
A profit factor of 1.48? Other robots show 3.
- 1.48 is after real spreads and swaps, and across 8 markets — not on one hand-picked best market.
- A high profit factor is cheap in a backtest: assume zero costs, or tune the settings on the same period.
- More than 1,400 trades over more than 6 years stand behind this figure. On 50 trades, statistics can say anything.
Isn’t a 53% win rate low?
- A win rate means nothing on its own. Martingale robots sometimes win 90% of the time and burn out on one loss.
- In LFP the average winning trade is larger than the average losing trade, and every trade’s risk is known in advance.
2023 was only +18%. Has the robot weakened?
- 2023 was a weaker year for the trend and breakout parts; for part of the year gold’s daily range fell to its lowest of the period. Even so, the year did not lose.
- 2024 and 2025, the years of the independent test period, were strong again (+90% and +79%).
- That is the point of diversification: a weak year for one logic can be a good year for another.
Could the robot’s edge disappear?
Yes — and we say so plainly.
- No edge in the market lasts forever. The mean-reversion part’s edge, for example, has narrowed in recent years, and we have measured it.
- Our answer: the portfolio holds 6 strategies so that one weakening does not take the system down. The watchdog stands down a market that no longer pays. Every 12 months a full review runs on new data, under the same strict rules — never by re-tuning on a few bad months.
Why was a strategy removed from the portfolio?
- Because the broker raised that market’s spread (the Nikkei index) about 100-fold and it was no longer profitable. Removing an ineffective part openly is a sign of a healthy process, not of a weak robot.
Why is there a minimum capital?
- The robot sizes each trade as a percentage of the account, and brokers have a minimum trade volume (0.01 lot). On a very small account most trades fall below that minimum, and the robot is no longer the portfolio that was tested.
- Instead of rounding the size up — which would mean more hidden risk than agreed — the trade is not placed, and we state the minimum capital plainly: about $2,000 on a standard-contract account.
Why did I lose after I started?
- Starting at the beginning of a drawdown is possible with any real system. The Brave style’s deepest historical drawdown was 20.6%, and the future can be worse.
- That is why the account guards are set in advance, and why we suggest starting with an amount whose 20% drawdown would not affect your life.
Check the numbers yourself
See every trading style’s result on your own capital in the LFP Console, and before buying, use the free compatibility test to see which markets your broker offers and what they cost.