Last spring my bot lost money on nine of fourteen days. Net for the two weeks: up 6%.
If you'd only looked at the daily P&L, you'd have called it broken. I nearly did.
Red days are the design, not the bug
My main trend system wins about 35% of its trades. That's normal for trend following. You take lots of small losses while waiting for the few moves that run for weeks, and those few pay for everything else.
Do the math on a simple version. Win 35 trades out of 100 at an average of +3R, where R is the amount you risk per trade. Lose the other 65 at -1R. That's +105R against -65R, so +40R net. A strong result, built on a system that is wrong almost two times out of three.
Now feel it from the inside. Most trades lose. Most days end red. Streaks of five or six losses show up regularly, and nothing is wrong when they do.
The gap nobody warns you about
Your brain doesn't track expectancy. It tracks the last few outcomes, and it treats losses as roughly twice as loud as gains. That's the standard finding from prospect theory work by Kahneman and Tversky, and I can confirm it from my own screen.
So a system that is working produces a steady stream of small pain and rare, delayed pleasure. Your gut reads that as failure. Your spreadsheet reads it as Tuesday.
That gap is where most people quit. Not after a blowup. After a quiet, boring, perfectly normal run of red days that the backtest had already shown them.
I know a trader who dropped a profitable system in week five, one trade before the move that would have paid for the whole quarter. The system didn't fail him. The feeling did.
What most people do about it
They start tweaking. Tighter stops so the losses hurt less. Extra filters to skip the bad ones. A target at 1R, because a quick win feels so good.
Every one of those changes buys comfort and pays for it in edge. Cut the winners short and the 35% system turns into a 55% system that makes less money. It feels better and earns less. I've done this, and I wouldn't recommend the experience.
Others go the opposite way and stare at the equity curve all day. Which just means feeling every red candle in high definition.
What actually helps
Judge the system on the sample, not the day. Before going live, write down the longest losing streak and the worst drawdown in your backtest, then add a buffer. If the backtest says eight losses in a row happened once, plan for ten.
Then make the review window longer than your emotional one. I look at results monthly and check the daily number only to confirm the bot is running. A single red day carries almost no information. Fifty trades carry some. Two hundred carry quite a lot.
And keep a one-line note of what the system is supposed to feel like. Mine says: "Most days are red. That is the cost of the winners." It sounds silly. I read it more often than I'd admit.
The practical takeaway
Open your backtest and find your win rate and your longest losing streak. Write both on a card where you can see them. When the live results start to feel awful, compare them to the card before you touch a single setting.
If the live numbers sit inside what the card says, nothing is broken. You're just feeling the price of the edge.
If you want to see how a low-win-rate system looks over years rather than days, our full backtest data and methodology are at v33systematic.com.