بناء نظام تداول واختباره: من الفكرة إلى الدليل

This is the last station in the survey, and it assembles everything the previous forty-three sheets built: the mechanics, the costs, the sizing arithmetic, the plan, the journal. A trading system is a set of rules a stranger could execute without asking you a single question — entries, exits, sizing, and the conditions under which you do nothing. Most trading ideas never survive contact with that sentence, and most that do are then killed by one of two documented failure modes: overfitting to the past, or abandoning the rules mid-drawdown. This sheet walks the whole road — idea, specification, historical test, forward test, live graduation — with both failure modes treated as the main events they are.

Rules a stranger could execute

Start with the specification test, because it is cheap and brutal. Take any trading idea you currently believe — say, “buy pullbacks in an uptrend” — and imagine handing it to a stranger who will trade it for a month with their own money, no questions allowed. They need answers immediately: which pair, which timeframe? What defines the trend — and what officially ends it? How deep must a pullback go before it counts, and what confirms the entry? Where is the stop, where is the target, what size? What happens when a news event lands mid-trade? Every question the stranger would ask marks a judgment gap — a place where, live, your mood would be making the decision while calling itself analysis.

Writing the answers down is what converts an idea into a system, and the exercise is worth doing even if you never trade the result. Most ideas die here, honestly and cheaply, the moment they are asked to be precise — which is a far better place to die than in your account three months from now. The ones that survive acquire a property no discretionary hunch ever has: they can be tested, because two people running the same specification on the same prices get the same trades.

Choosing a style honestly

Before any rules are written, a prior question deserves an honest answer: what kind of trading does your actual life have room for? The three broad styles differ mainly in holding period and attention demanded, and the differences are practical, not aspirational:

Read the table as a constraint, not a menu of earnings. An employed person choosing ▲scalping has chosen a second job they cannot attend; their system will be executed badly not from weakness but from arithmetic — the attention it specifies does not exist in their week. ▲Day trading suits a free, regular daily window and a temperament that can close flat every evening. ▲Swing trading fits around employment but holds positions through nights and weekends, which means financing costs and gap risk are part of the specification, not footnotes to it. No style is the profitable one. Each is a different bill, paid in a different currency — attention, costs, or overnight exposure — and the right choice is the bill you can actually afford to keep paying.

From idea to ruleset

A complete specification has four components, and beginners reliably over-build the first while neglecting the other three. Entry rules: the conditions that must all be true to open a position. Exit rules: the stop and the target, both placed at entry — fixing the ▲risk-reward ratio inside the specification rather than leaving it to the moment. Sizing: a fixed-fraction rule, exactly as the plan sheet specifies. And no-trade conditions: the circumstances under which a valid-looking entry is refused — around major scheduled news, when the spread exceeds a stated ceiling, outside the system's session, or after the daily loss limit.

Where do the entry ideas themselves come from? From everything the survey already covered — a moving-average structure from the technical region, a session-timing observation, a pattern your journal keeps surfacing. The source matters less than beginners assume, because the specification and the testing do the real sorting: a mediocre idea specified precisely and tested honestly is worth more than a brilliant intuition that cannot survive being written down. Aim for few moving parts. Every additional condition is another parameter to overfit, and the systems that survive forward testing tend to embarrass their builders with how plain they are.

Here is what the shape looks like in practice — illustrative only, chosen for clarity rather than merit:

• Market and timeframe: EUR/USD, daily chart, evaluated once at 22:00 platform time.
• Trend filter: price above its 50-day moving average — below it, this long-only system stays flat.
• Entry: a close back above the 20-day average after at least three days below it, while the trend filter holds.
• Stop: below the lowest low of the pullback, capped at 60 pips. If the cap would be exceeded, no trade.
• Target: 1.5× the stop distance. Both orders placed with the entry.
• Sizing: 0.5% of account risked per trade — on a $4,000 account that is $20; with a 50-pip stop, $20 ÷ (50 × $10) = 0.04 lots.
• No-trade: within 24 hours of a rate decision on either side of the pair; spread above 2 pips; an existing open euro position.

Run the symmetry check on it like any worked example in this survey: at 0.04 lots a winning trade pays about $30 (75 pips × $0.40 per pip) and a losing one costs about $20 — and nothing in the specification tells you how often each occurs. That frequency question is the entire remaining content of this sheet, and no amount of specification answers it. Only testing does.

Backtesting without fooling yourself

A backtest replays the specification over historical prices to see what it would have done. It is genuinely useful and systematically flattering, and the flattery has three standard mechanisms. The first is hindsight. Scrolling history, your eye already knows where price went, and it nudges every borderline judgment the way the chart resolved — setups look obvious that live, on the right edge of an unfinished chart, you would have missed or doubted. The discipline is to test the written rules exactly as the stranger would, bar by bar, deciding before revealing what came next.

The second mechanism is ignored costs. Suppose the historical run produces 200 trades and a paper profit of $600 at 0.2 lots. At a one-pip spread, each round trip cost about $2 that the replay never charged: 200 × $2 = $400, and the $600 edge is suddenly $200 — before slippage on entries and gaps through stops, which historical bars hide entirely and which only ever subtract. A backtest that has not charged realistic costs has not produced a result; it has produced an upper bound. The symmetric point holds too: costs shrink every win and deepen every loss on the same schedule, so the thinner the paper edge, the more completely the bill erases it.

The third mechanism is the deepest: overfitting. Tune any parameter to improve the historical result — try 18- and 22-day averages, nudge the stop cap, add a filter that deletes 2019's worst month — and the past will always reward you, because the past holds still while you aim. Each tuning pass fits the rules more tightly to that particular history's noise, and noise does not repeat. The tell is fragility: if moving one average from 20 to 25 days flips the system from winner to loser, you have not found an edge, you have memorised a dataset. A rule added to remove one historical losing trade is the purest form of the disease — and it always improves the backtest.

Held to those standards, a backtest answers one question well: should this idea be rejected, or tested forward? It rejects cheaply and honestly — a specification that loses on generous historical assumptions needs no further weeks of your life. What it can never do is prove the system works, because the only evidence it consults is the one dataset your rules have already seen.

The forward test: a fixed exam, criteria first

The forward test runs the specification on a ▲demo account, in real time, on prices neither you nor the rules have ever seen. This is the real exam, and its integrity comes entirely from one habit: the success criteria are written down before the test begins. Decided afterwards, criteria bend around whatever happened — a poor result becomes “promising,” a violation becomes “an exception” — and the exam silently turns into one more backtest, this time fitted to your hopes instead of to history.

1. Fix the length first: a minimum trade count (30 to 50) rather than a calendar window, so a quiet market extends the test instead of ending it early.
2. Pre-register the thresholds: the expectancy the journal must show to pass, the maximum drawdown in R that fails the test immediately, and the plan-adherence rate below which the result is void — a test you didn't follow is a test of nothing.
3. Run it through the full journal — all ten fields, weekly reviews included. The forward test examines the operator as much as the rules.
4. Score it against the written criteria at the end, exactly as the stranger would — no partial credit, no “but for those two trades.”

Be honest about the instrument's limits in both directions. Demo flatters execution — fills come easier and stops gap less than they will live — and trading without consequence is psychologically easier than it will ever be again. And a pass can also simply be luck: 40 trades is a small sample, wide enough for a mediocre system to look good and for a decent one to look poor. The forward test does not remove uncertainty. It shrinks it to a size that a small live trial can afford to carry — which is all any test can honestly claim.

Live, small, and monitored

Graduation, if it comes, is deliberately anticlimactic: the smallest size your sizing rule allows, with the journal running and one new column added — execution quality. Live fills versus demo fills, real slippage, the spread actually paid at your trading hour. The first weeks live are not for earning; they are for measuring the gap between the rehearsal and the stage, and for checking that the forward-test numbers survive contact with real friction.

Two sets of rules govern everything after that, and both are written before going live. Graduation criteria: size increases only on schedule — after a defined number of live trades with expectancy intact — never as a reward for a good week. And quit rules: the drawdown in R, or the adherence collapse, at which the system stops trading, full stop. Quit rules deserve respect, not embarrassment. A system retired by its own rule is a success of process, and the journal it leaves behind is the seed of the next, better specification.

The quit rule exists because of the second great failure mode: abandonment mid-drawdown. Every system, including genuinely sound ones, spends long stretches under water — the drawdown sheet's arithmetic makes that unavoidable. The moment of deepest drawdown is when the urge to “just adjust one rule” peaks, and it is precisely the moment your judgment about the system is worst, for every reason the psychology sheets documented. So the discipline is binary by design: between reviews, the system runs as written or it stops by rule. Redesign happens after a stop, from the journal, calm — never inside the hole. Most retail systems are not defeated by their markets. They are dismantled by their own operators, one mid-drawdown improvement at a time.

And that is where the survey ends — deliberately without a finish line. A specification, an honest pair of tests, a small live trial, a journal that audits all three, and rules for stopping: that is the complete toolkit this curriculum can give you. What it cannot give you is an edge, and after forty-four sheets you know exactly why nobody honest will promise you one.