How an AI Earns $5,000 in 7 Days, and Why It Still Fails Every Second Prop Challenge

I love payouts from prop firms. But I hate the challenges. So I built a system that cracks the challenges for me on autopilot. One after the other.

Here is the strategy at its core:

$10,000 turned into $58,395 in just under two years. 1,035 trades, 54% winning trades, Sharpe 1.22.

Nice curve. But a challenge is not a two-year race. It is a sprint, and to understand why this strategy fits it, you first need to know how a challenge is actually passed.

How a strategy passes a challenge

A prop firm gives you a $50,000 account and sets two lines. Above +$5,000 you have passed. Below −$3,000 you have failed. Whoever touches a line first decides the challenge.

So the job is simple: get from 0 to +$5,000 as fast as possible, without ever touching −$3,000 on the way.

Here is how that looks over time:

• Day 0: The challenge starts at 0. The strategy waits for its first signal.
• Days 1 to 2: The market starts to move. The account dips into the red, because positions are being added. This is normal and planned. The dip stays far away from the −$3,000 line.
• Day 2: The market turns back for a moment. All positions close together at the profit target. The first gain is banked, and the account is in the green.
• Days 3 to 6: The same game again: a dip, a recovery, the next gain banked. Every finished cycle is a step up the staircase.
• Day 7: The last cycle pushes the account over +$5,000. The challenge is passed, and the plan is to stop trading right there.

The red dashed line shows the other case. If one dip goes deeper than the strategy can bear, the account touches the −$3,000 line before the next gain is banked. That is a failed challenge.

You can see what matters: a challenge strategy only has to make the way from 0 to +$5,000 once. After that, a stable, slow strategy can take over, one that is optimized for continuous payouts. Right now we want to be fast, because every extra week in the challenge may cost us $500 in payouts.

So a challenge does not only cost the fee. You also pay with time, because every week in the challenge is a week without payouts.

That is why it is worth aiming to pass a challenge within 7 days. Even if you have to try three times in a row before one passes. After that the payouts start, and they make up for all the money you spent.

What the strategy does inside each of these steps, and how you can have your own version built in an hour, is what I show you on the next pages.

First the bad news

Before I sell you anything, let me tell you what this is not.

It is not a money machine. Almost every second simulated challenge failed, 44% to be exact, because the $3,000 limit was blown. There is no guarantee, and I have no interest in promising you one.

If you are looking for a sure 100%, stop reading here. If you want to understand how to approach a challenge with a system instead of nerves, and you can accept a rate of around 54% with clear rules, it is worth your time.

Why do I say this right at the start? Because the number is honest, and because you can check it yourself. I will show you how.

How the idea was born: The trend flips

Every good strategy starts with an observation every trader knows, and this one I had stared at for years.

An uptrend is a staircase. Price rises, pulls back a little, rises further. Every intermediate low is higher than the one before. As long as that holds, the trend is intact.

Then it happens: price breaks below the last intermediate low.

Traders call this a Change of Character, ChoCh for short. The character of the market has changed. The staircase suddenly leads downward.

The strategy looks for exactly this moment, in both directions. A swing indicator (period 6) marks the swing lows and highs. As soon as a candle closes below the last swing low, it goes short. As soon as a candle closes above the last swing high, it goes long.

The twist that blows the minds

Now comes the problem. A single ChoCh is often a false alarm. Price flips briefly and then keeps climbing. I used to set a stop-loss here, get stopped out and be annoyed.

The AI's idea was different:

Markets keep changing direction. After a ChoCh down comes a ChoCh up, then down again. Every one of them is a signal.

So the strategy does not bet once. It adds: every ChoCh opens another position in its direction, and the positions stay open.

1. First ChoCh (down): Price closes below the last swing low. Short 1, small size.
2. Price reverses: It breaks the last swing high, a ChoCh up. Long 2 is added, and the short stays open.
3. And again: ChoCh down, Short 3. ChoCh up, Long 4. With every position the size gets bigger.

The advantage: the large, late positions sit in the direction of the latest swing. Once the package is in profit together, even a small move in that direction is enough to reach the target, and all positions are closed at once.

I saw right away where this gets dangerous. Growing sizes: that is a martingale structure.

I almost stopped right there.

The trick: stop loss for the package, not single positions

What the AI did next looks illogical at first: it removed all single-position stop-losses.

The reason is quickly explained. In a growing series, a single stop would trigger the old, small positions first and let the large ones run. I would lose exactly in the wrong places.

Instead, all positions are managed as one package (a "basket"). The package has three rules:

• Profit target: When the package reaches the target profit, all positions are closed. The profit is banked and a new cycle starts with the small starting size.
• Emergency brake: If the package falls to a hard loss limit, everything is closed as well. The loss per cycle is capped.
• Safety line: If the package gets very large (many open positions) and is in profit, a retreat line is armed. If the profit later falls back to it, everything is closed before a good package turns into a loss.

And one more thing: the AI was not asked to find maximum profit, but a good ratio of profit to drawdown. It trained on a short window (so the name says: 6montrain), and the result was then checked on the remaining data.

That is the whole strategy. One idea, one trick, three rules. No wall of indicators, no gut feeling.

The proof: 250 challenges, each started on a different day

Anyone can show a nice equity curve. I was interested in the question that matters to you: How often do you pass a real challenge with it, and how fast?

I scaled the curve up to a $50,000 account. Factor 4 means the profits and losses of the $10,000 backtest were multiplied by four, as if the strategy had traded with four times the position size (scaled linearly). Then I started a challenge (+$5,000 target, $3,000 drawdown limit) on every possible start day. 250 start days, 250 challenges.

136 of 250 challenges were passed (54%).

And when it worked, it was fast: the median is 7 days. Half of the successful attempts were done after one week, almost 90% after two weeks.

Seven days for $5,000. This is the moment you understand why this kind of strategy is so fascinating.

When it doesn't work, and what that means for you

Almost all of the other half, 44% of all starts, blew the $3,000 limit. Why? Because the strategy is a martingale variant. Most of the time it runs quietly, and every now and then there is a drawdown deep enough to end a challenge.

The lower strip shows every start day. Green: passed. Red: failed. There are long green phases and there are red blocks. The strategy is not just "sometimes good, sometimes bad": it depends on when you start whether the challenge runs into a drawdown or sails past it.

So: this is a tool for short goals. A challenge only wants 10% in a short time, and that is exactly what the ChoCh accumulator is built for. If an attempt fails, it costs you the challenge fee. If it succeeds, you have in a week what others do not manage in months.

When to start: Right before the big news

If it depends on when you start, the obvious question is: is there a start day with the best odds of passing right away?

The best time is just before big market events. Think of the FOMC decision, the CPI report, the NFP jobs report, or the end of a quarter, when the gold roll-over at the COMEX takes place.

Before the events, a simpler question: does the weekday of the start matter?

Thursday and Tuesday look best, Monday and Wednesday worst. But every bar has only 38 to 57 starts, and the gap between the best and the worst day is about what chance alone produces. I would not plan a start around a weekday.

And what if you activate the strategy right around the event? For each of the three I compared the 24 hours before and the 24 hours after the release and kept the better window:

The sample size is rather small, with only 6 to 11 starts per window, so treat these numbers as an indicator, not as proof. Still, I will run a prop firm account with this strategy right before the next CPI release.

So instead of starting on a random Monday, you can plan the start.

What is the best position size to pass quickly?

So far I have calculated with factor 4. But the lot factor is a dial, and the right setting depends on how much time your prop firm gives you. I ran the factor from 1 to 20 and searched for the best value for each time limit:

The rule behind it: little time demands a lot of risk, a lot of time allows little risk. With factor 1 and 120 days of time, two out of three challenges pass. If you want to be done in a week, you have to turn the position size up so high that the challenge becomes practically a coin flip.

This is a statistical evaluation. It shows an edge in the market, but no guarantee.

How to synchronize the passing lot factor with the lot factor for a funded strategy?

The idea: once the challenge is passed, I switch to a different, less risky strategy for the funded account. It was also built on the minimum lot size of 0.01, so it has to be scaled up to fit the account size as well.

That gives two factors to choose: F1 for passing the challenge and F2 for generating payouts.

The following simulation shows how many payouts minus challenge fees can be reached within 6 months. If a passing or funded account blows up, a new one is bought right away.

Your move: Build your own version, today

Now it is your turn. This is the shortcut. Copy the prompt, give it to the AI-Backbone builder agent, and in about an hour you have your own version with equity curve, drawdown and prop pass rate.

Build a strategy "choch_accumulator" for XAUUSD and save it as the current strategy in the workshop.

STRUCTURE
- Use ChoCh indicator with a swing period parameter that can be optimized.
- ChoCh DOWN: a close breaks the last confirmed swing low -> SHORT.
- ChoCh UP: a close breaks the last confirmed swing high -> LONG.
- For every confirmed new ChoCh in the same direction, add one more
  position (accumulation). All open positions belong to one basket and
  are only closed together.

POSITION SIZING (two-stage martingale)
- First position of a cycle: use parameter position_size (default and minimal: 0.01 lot).
- While open_count < parameter k1: next size = previous size * f1.
- From open_count >= parameter k1: next size = previous size * f2.
- Set a hard cap on position size and on the number of open positions as parameters that can be optimized.

RISK (no stop-loss per position, basket exits only, checked on every bar)
- profit_target_usd: close ALL positions when the basket PnL >= target.
- basket_stop_usd: close ALL positions when the basket PnL <= -stop.
- use parameters break_even_pos_count + be_offset_usd: as soon as MORE than N positions
  are open and the basket is in profit, a latch is armed. After that, close
  ALL positions as soon as the basket PnL falls back to <= be_offset_usd.
- After every basket exit, a new cycle starts with the base size.
- use parameter breakeven_max_pos: if this number of open positions is reached, close all positions as soon as pnl>0 is reached

OPTIMIZATION (ranges)
- f1: 1.0 to 5.0, f2: 1.0 to 5.0, k1: 1 to 20
- profit_target_usd: 1 to 10000, basket_stop_usd: 10 to 10000
- break_even_pos_count: 1 to 50, be_offset_usd: 0 to 1000

Here is how to proceed:

1. Create a free subscription at ai-backbone.com/subscribe.
2. Copy the prompt and send it to the AI-Backbone builder agent.
3. Check the result: equity curve, drawdown and the prop pass rate from step 4 of the prompt.
4. Use the MCP connector with your Claude or Codex instance to adapt the optimization and the idea of this article to your own prop firm accounts.

Conclusion

One idea (the trend flips). One trick (add instead of betting once). One safeguard (risk in the package instead of the single position). And an AI built all of it in a short time.

Build it. Improve it.

P.S. The one who is in control, can make money without luck.