If your account has a limit on open positions — and every account does — there are days when more marks arrive than fit. What decides which ones get in is usually chance. We set out to measure how much that chance weighs, and it weighs more than almost everything we were comparing.
The same strategy, the same data and the same rules, run twenty-five times. Results between 185.73 and 304.78. A hundred and nineteen points of difference without touching a single rule.
A system generates marks. An account has room for a certain number of positions at once: ten, let's say. On quiet days there's no problem. On days when fifteen marks arrive and there are ten slots, someone has to decide which five are left out.
In most programs, nobody makes that decision. It's made by the order in which the symbols appear in the list, which is almost always alphabetical. In other words: whether a trade gets in or not can depend on the company being called Adobe and not Zoom.
And it isn't a rare case. In our system's backtest, out of 11,988 marks generated over sixteen and a half years, with ten simultaneous positions only 1,488 actually get traded. 87.6 % never get in, not because they're bad, but because when they arrived there was no free slot. With that proportion, the tie-break isn't a detail: it's a large part of what ends up happening.
We ran the same configuration twenty-five times, changing just one thing: the order in which ties between same-day marks are resolved. Not a single entry, exit or risk rule touched.
| Over 25 runs identical except for the tie-break | Result (R) |
|---|---|
| The worst | 185,73 |
| The median | 251,66 |
| The best | 304,78 |
| Range from end to end | 119,05 |
| Standard deviation | 30,96 |
The consequence is uncomfortable, and it's the one that's needed: if you compare two configurations by running each one just once, and the difference between them is thirty points, you haven't measured anything. That difference fits entirely inside the randomness of the allocation.
We checked it with our own comparisons. We had tested fifteen variations of the system with one run each, and eleven seemed to stand out from the original. When we repeated each one twenty-five times and compared medians, only four were still different. Ten were within the noise, and one pointed in exactly the opposite direction from what the single run had suggested.
Because the scarce resource isn't the mark: it's the slot. When a trade gets in, it occupies a place for weeks. Everything that arrives in the meantime and doesn't fit disappears — and doesn't come back. Each tie-break doesn't change one trade: it changes the whole chain of what could be traded afterwards.
That's why the effect is so large. It isn't twenty small independent differences adding up: it's one initial difference spreading across sixteen years of chained decisions.
And there's a practical consequence you can see immediately: how long each trade stays open matters as much as which ones you choose, because time open is what uses up the slot.
To answer the question of whether a strategy works or you've been lucky, which almost nobody answers. The practical answer is simple: run the same test several times changing what shouldn't matter, and see how much the result moves. That movement is your noise margin. Any improvement smaller than that isn't an improvement: it's a lucky run.
It's especially useful before paying for a prop firm challenge. If your system, repeated twenty times on the same history, gives results ranging from losing to winning, you already know the result of the real challenge will depend quite a lot on chance — and that's useful information before putting in the money, not after.
And it changes what's worth looking at when you compare two versions: not the number, but the number and its spread. A version that gives the same on average but varies half as much is the better version, even if the headline doesn't say so.
It doesn't say historical results are useless. It says a result that hasn't been repeated doesn't let you compare two similar things. It doesn't say twenty-five repetitions are enough: they're the ones we did, and with more the margin narrows. It doesn't say this margin is your system's — it depends on how many marks you generate, how many slots you have and how long your trades last, so it has to be measured on your own. And it says nothing about what anyone should do with their money.
With a test you can do yourself, and it takes an afternoon: repeat your backtest changing the order of your symbols. If the result moves a lot, your position limit has more say than your entry criterion, and you know where to work.
And with a different way of reading any number: before asking how much it made, ask how many times it was run. If the answer is once, what you're looking at is a possibility, not a result.
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 a high win rate isn't good news. They are 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, 10 simultaneous positions, exit at 3 R. 25 runs identical except for the tie-break between same-day marks. Measured over 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 same-day marks, and the median is published. Educational content, not financial advice. We don't sell signals or manage third-party capital. Familia FVR · Sophronepsis.