It's the first number shown and the most reassuring one. It's also the easiest to raise without improving anything. We put figures on it by tightening the filter of our own system: the win rate moved and the money didn't.
We required our system to meet all six conditions instead of five. The win rate rose —slightly, half a point— and the money stayed the same: 247.80 R against 251.66, a difference smaller than how much the system itself varies between two runs. What did change: it traded two-thirds fewer times.
The win rate tells you how many trades end up positive. It doesn't say how much is made on them or how much is lost on the others. They're two different questions and only one gets answered.
And it can be raised at will without touching the quality of anything: just close the winners early and let the losers run. Every trade closed early with a small gain adds a hit. Every loss left open hoping it comes back doesn't count as a miss yet. With those two habits, the win rate goes up and the account goes down.
That's why, when a win rate is very high, the first thing worth asking isn't how it was achieved: it's how much it loses when it's wrong. A system that's right 85 % of the time and loses on each miss ten times what it makes on each hit loses money. A system that's right 40 % of the time and makes three times what it risks, makes money.
Our system checks six conditions before entering and trades when at least five are met. The idea of tightening it to six seems obvious: more demanding, better entries, more hits. It was also going around in-house as something taken for granted, so we went to measure it.
| Requirement | Result (R) | Win rate | Profit factor | Maximum drawdown | Variation between runs |
|---|---|---|---|---|---|
| 5 of 6 (what it trades today) | 251,66 | 40,7 % | 1,357 | −43,13 | 30,96 |
| 6 of 6 | 247,80 | 41,2 % | 1,425 | −32,77 | 18,10 |
Tightening didn't change the money. The difference in result is 3.86 R. And the same system, run twenty-five times changing only the order in which it handles the marks that arrive on the same day, swings between 18 and 31 R from one run to the next. The difference fits entirely inside that swing: with this data we can't say that one version beats the other, in either direction.
What does change, and a lot, is how many times it trades. And the win rate rises half a point. There's the whole report in one line: more demanding, slightly higher win rate, the same money. If the win rate measured quality, the strict version would be better. It isn't: it's the same one, trading a third as often.
The answer lies in how the marks are distributed, and it's not the one we expected. Of the 11,988 marks in the backtest, 8,247 meet five conditions and 3,741 meet all six. Requiring all six doesn't remove a minority: it removes 68.8 %. Two out of three.
And that's the explanation for why the money doesn't move. If you remove the limit on open positions and let the two groups run separately, they return almost the same per trade: 0.189 for the five-condition ones, 0.185 for the six-condition ones. The sixth requirement doesn't tell a good mark from a bad one. That's why removing two-thirds of the marks doesn't remove the bad ones: it removes a bit of everything in the same proportion, and the result stays where it was.
And there's a second effect, which explains why it doesn't improve either. A normal account can only have a handful of positions open at once, so most marks never get traded at all: with ten positions, of the 11,988 only 1,485 are traded. When the filter removes an entry, there's almost always another one just as good waiting to take that slot. Tightening doesn't leave better entries: it leaves the same ones, less often. Fewer entries is not the same as better entries.
The win rate didn't fall: it rose half a point, from 40.7 % to 41.2 %. Half a point means nothing either with this swing between runs, and that's why we build nothing on top of it. We note it because in the first version of this report we said it had fallen, and that wasn't true.
Correction — 16 August 2026
The first version of this report said that «10,824 of the 11,988 marks already met all six conditions, 90 %», and concluded that the filter barely separated anything. It was false. That 10,824 was the total number of entries from another run, read as if it were a split. Counted correctly, those that meet all six are 3,741 — 31.2 % —, so the filter does separate, and a lot: it removes two out of every three marks. And that correction fell short: we said that «the report's conclusion doesn't change (requiring all six comes out worse in all five columns)», and it did change. It's explained below, in the 17 August correction. We leave it written here, exactly as it was published, instead of changing it quietly.
Second correction — 17 August 2026
Yesterday's correction fixed the split of marks and stated that «the report's conclusion doesn't change». It did change. The «6 of 6» row of the table had been measured on the wrong group —the same 10,824-mark error we corrected yesterday—, and we fixed the explanation while leaving standing the figures that came from it. Measuring the correct group, «6 of 6» doesn't give 192.14 R but 247.80, and the 24 % drop we announced doesn't exist: it's 3.86 R of difference, noise. Nor is it «worse in all five columns». We've redone the table and the reasoning that hung from it. The underlying idea —that the win rate doesn't measure quality— holds, and with a cleaner example than the one we had.
To stop using the win rate as a measure of quality. It's a consequence of where you put the exit, not a property of how good your entry is. A close target is reached more often and raises the win rate, leaving less on each hit; a distant one does the opposite. The system hasn't changed: where it closes has. Ours, with a target at 3 R, is right 40.7 % of the time.
The measure that does compare two systems is what each trade makes on average, counting the losers. If you prefer a single figure, the profit factor: how much comes in for every euro that goes out. An 85 % win rate, on its own, says nothing yet.
And to distrust the «more demanding is better» intuition. That reasoning sounds so good that it's almost never checked. Here, checking it avoided changing the system for nothing: we'd have traded a third as often only to end up in the same place.
It doesn't say tightening a filter is worse, nor that it's better: it says that in this system and with this data it doesn't show up in the money, and the reason is concrete —the marks that pass the filter return practically the same as those that don't—. In a system where the filter did separate good from bad, the result would be different. Nor does it say that the small differences in the table (the profit factor, the drawdown) mean anything: with that variation between runs, we don't take them as valid. It doesn't say a high win rate is suspicious in itself; it says that on its own, without knowing how much is lost on the misses, it tells you nothing. And it doesn't say what anyone should do with their money.
With two questions for any result you're shown: how much do you lose when you're wrong?, and where is the exit? With those two, a win rate starts to mean something. Without them, it's decoration.
And with a check you can run on your own strategy: split your marks into those that pass the filter you're thinking of adding and those that don't, and compare what each trade makes on average in the two groups. If the two numbers are similar, that filter isn't telling good from bad: it's only going to take trades away from you. Counting how many pass is no use — it may remove many or few, and in neither case does it tell you whether the ones it removes were worse.
If you want to keep pulling the thread
The other two reports in the series: how much history you need before trusting a strategy and why the same test repeated twenty-five times gives twenty-five different results. They're all at sophronepsis.com/informes.html.
And if you want to learn to look at this on your own, the six-day walkthrough is at app.sophronepsis.com/empieza.
A report is a snapshot of one day. Inside the platform the measurements are redone when new data arrives, every strategy comes with its test alongside it, and the mentor answers whatever you ask, at any hour. sophronepsis.com/mentoria.html
This is research, not a sermon. If you find a flaw in the method or in the numbers, write to us at hola@sophronepsis.com and we'll correct it in public. Sophronepsis · Wait. Observe. Execute.
Backtest: 11,988 marks generated on 148 stocks and 16.6 years of daily data, exit at 3 R, 25 shuffles per configuration and median published. Measured on 148 stocks and 16.6 years of daily data, with costs of 0.05 % commission and 0.05 % slippage, and without broker costs for holding the position open overnight. Each configuration is run 25 times shuffling the tie-break between marks on the same day, and the median is published. Educational content, not financial advice. We don't sell signals or manage third-party capital. Familia FVR · Sophronepsis.