1. Write the rules down before you test
Open a file. Write what the strategy does, what you are about to test, and what result would make you abandon it. Date it. That is the whole practice, and it is the highest-value habit in this course.
It works because of what happens otherwise. Without a written record, a disappointing result becomes an invitation to adjust something and try again, and after a dozen rounds of that you have searched a large space without ever deciding to. You will also not remember how many things you tried, which is the number the previous lessons said you need.
Writing the abandonment criterion first is the part people skip and the part that matters. Afterwards, every failure has an explanation available.
2. Hold data back and do not look at it
Split your history. Develop on the earlier portion, and keep a later portion sealed. Do not chart it, do not test on it, do not glance at it. When you have a finished strategy, run it once on the held-back data.
The rule is that you get one look. If you test on it, adjust the strategy, and test again, it is no longer held-back data. It has become part of the search, and you have quietly spent the only clean evidence you had.
3. Count your variations
Keep a running count of every configuration you test, including the manual ones and the abandoned ones. Optimiser runs count in full. If you swept 400 combinations, that is 400, not one.
Keep the results too, not just the winner. The spread of outcomes across everything you tried is what tells you whether the best one stood out or merely came first, and it is the input an honest overfitting analysis needs.
4. Check costs before anything else
Do this second, right after the idea and before any real work, because it is fast and it disqualifies a lot. Compare the move you are trying to capture with the round-trip cost of capturing it at your broker’s actual figures. Re-run at double the cost you assumed and see what survives. Lesson two has the arithmetic.
5. Forward-test on a demo
Run it on a demo account, unchanged, for long enough to accumulate a meaningful number of trades. This is slow and it is the only evidence generated on data that did not exist when you built the strategy.
Two cautions. A demo does not reproduce real fills, so treat it as a test of the logic and the plumbing rather than a forecast of returns. And if you change the strategy during the forward test, the test restarts. It does not average.
6. Size so that being wrong is survivable
Position sizing that maximises expected growth also maximises the chance of being wiped out, because it assumes you have measured your edge correctly and you have not. You have an estimate from a limited sample.
Size smaller than the arithmetic suggests. Then check what happens to your account under a losing streak of the length the previous lesson described, and confirm you would still be trading afterwards. If a normal bad patch would end you, the size is wrong regardless of what the expected value says.
Rules and abandonment criteria written down and dated. Data held back and looked at once. Every variation counted, results kept. Costs checked at real figures and at double. Forward-tested unchanged. Sized so that an ordinary losing streak is survivable.
What none of this gets you
Doing all of it does not mean the strategy will work. It means you have removed the ways of fooling yourself that are removable, and what remains is genuine uncertainty about the future, which nobody can test away.
That is the honest ceiling on this entire subject. Validation can find evidence that a strategy is not what you think it is. Nothing can establish that it will make money. Anyone offering you the second thing is selling something.
If you want the statistical versions of these checks run on a trade log you already have, that is what this site does, and you can look at a sample report without an account. If you would rather do it yourself with the methods above, that is a completely reasonable outcome for this course.