Process vs outcome

15:10 ET (03:10 ICT, winter). Your rejection method just printed its sixth loss in a row. Your stomach says the method is broken. Arithmetic says something different: a method that wins 40% of the time runs into six straight losses in most 100-trade samples. Your stomach is not a statistical instrument. This lesson gives you one, and answers a single question: when is a result allowed to change your method?

Key terms

Process (quy trình): the part of trading you control (the plan, the execution against it and the review) Definition. Outcome (kết quả): the P&L of a trade or a period, which mixes your edge with luck Definition.

Variance (phương sai): how far results spread around their average. Over a small number of trades the spread is much larger than any realistic edge, so an outcome mostly reports variance Definition.

Term you need: expectancy in R (kỳ vọng theo R). The average result per trade in units of your risk: win rate × average win − loss rate × average loss. M6.08 applies it to entries; RSK1.01 in the Risk course owns it.

One method, one ordering of its trades

How it works

One trade is noise

A trade is one draw from a distribution you only partly know. Your edge, if you have one, lives in that distribution’s average, and an average needs many draws Definition. Judging a method on a streak is like judging a coin on three flips: the sample is too small to hold the information you are trying to pull out of it.

Streaks are arithmetic

At a 40% win rate the arithmetic, assuming independent trades with a fixed win rate, is plain Definition:

• six losses in a given six-trade window: 0.6⁶ ≈ 4.7%;
• at least one run of six or more losses somewhere in 100 trades: about 87%; in 50 trades, about 63%;
• two or fewer wins in the first ten trades: about 17% (exactly two: 12%).

So the “broken” streak in the hook is what a working 40% method produces in most samples. RSK1.02 in the Risk course owns the streak maths and its simulator; here you only need the conclusion: a streak, on its own, is not evidence against a method.

What a sample can tell you

Put an interval on the average and the size of the problem shows Definition. For the method in the figure (spread about 1.7R per trade), the 95% interval on the measured expectancy is roughly:

At 20 trades a +0.4R method and a −0.3R method look the same. Even at 100 trades only a large edge separates from zero: an edge of +0.2R needs about 280 trades before an interval centred on it clears zero, and +0.1R about 1,100. EV3.01 owns interval width and power; M7.10 applies it to validation.

The house ladder

Because of those widths, our house rule fixes when a method may be judged, so the decision is made before the streak arrives House rule:

• Under 20 trades: noise. Change nothing, conclude nothing.
• 20 to 50: log and watch; still no method change.
• 50 to 100: the shape appears: realised win rate, average win and loss, drawdown character.
• 100 or more: a method review is allowed, one variable at a time (M7.03), and the verdict comes with its interval, not a feeling.

The thresholds are a policy, not a measured cutoff. 100 trades does not prove an edge; it is the point where a review stops being pure noise for edges of the size retail methods usually plan for.

Two failures, one rule

The ladder blocks two opposite errors. One: abandoning a working method in a normal drawdown (quitting at trade 20, the M6.08 example). Two: trusting a losing method after a lucky run. Judging a decision by its outcome rather than by the information available when it was made is a well-documented bias in decision experiments MeasuredOther market; that it drives traders’ method changes is plausible but not measured on traders PlausibleOther market. M7.05 owns the bias list.

Separate the scoreboards

Score execution daily: adherence to the plan, in percent (M7.03). Score the method only at the ladder’s review points. A red day never rewrites the method; a green month never excuses red adherence House rule.

Worked example

Synthetic trade logs, no real session. Two traders run the same rejection method: 40% wins, +2.5R per win, −1R per loss, planned expectancy +0.4R. 1R is $150 (3 MNQ on a 25-point stop: 3 × 25 × $2).

After 18 trades both have 4 wins and 14 losses: −4R, −$600, −0.22R per trade. The 95% interval on that average runs from about −0.91R to +0.47R. It contains the planned +0.4R and it contains a losing method. The sample cannot tell them apart, and a start this bad (4 or fewer wins in 18) happens to a working 40% method about 9% of the time.

• Trader A reads −4R as “broken” and switches to breakout entries mid-month. By trade 60 A has 18 trades on one method and 42 on another, no usable expectancy estimate for either, and less confidence than before.
• Trader B checks adherence (94%: the execution was clean) and keeps going. By trade 60 B has 23 wins: +20.5R, +0.34R per trade. The interval is still about −0.09R to +0.78R. B does not know the method works; B has an estimate that is consistent with the plan and a clear next checkpoint at 100 trades.

Both traders were paid the same variance. Only B collected the information.

The mirror and the failure

The mirror: a method with a 25% win rate at the same +2.5R / −1R loses 0.125R per trade. It still wins 6 or more of its first 15 trades about 15% of the time, which shows as +6R, +0.4R per trade, the same number as the good method’s plan. A trader who scales up on that run makes Trader A’s mistake in the other direction. The ladder answers both the same way: not yet.

Where the ladder fails: it assumes the method stays the same while the sample builds. If conditions change (a volatility regime, a rule you quietly dropped), 100 trades can mix two different methods, and the interval describes neither. Log the conditions with each trade so a review can split the sample (M7.10).

Prop-firm note. An evaluation’s drawdown limit can end a sample long before the ladder’s review point. Size so that a normal losing run (six or more losses is common) fits inside the limit with room left; M7.08 covers drawdown rules, and the Prop Firm course covers firm-specific terms.

Common mistakes

Common mistake: “The method stopped working” after 12 trades. Twelve trades cannot separate a broken method from normal variance. The ladder’s answer below 50 trades is not “no, it works”; it is “the question cannot be answered yet.”

Common mistake: Treating a hot start as proof. A +0.4R average over 15 trades is something a losing method prints by chance about one time in seven. Read the interval, not the average.

Common mistake: Changing the entry and the stop at the same review. If expectancy then moves, you cannot tell which change did it. One variable per review, before and after logged (M7.03).

Drill: put an interval on your own sample

Take your last 20 to 50 journaled trades (sim is fine), with each result in R.

1. Compute the average R per trade and the standard deviation of the R results.
2. Compute the 95% interval: average ± 1.96 × standard deviation ÷ √n.
3. Write one sentence: does the interval contain zero? Does it contain the expectancy your plan assumed? What does the ladder allow you to change now?

Recap

• A streak or a bad month is mostly variance: at a 40% win rate, a six-loss run shows up in about 87% of 100-trade samples, and two or fewer wins in the first ten happens about 17% of the time.
• Read expectancy with an interval: about ±0.75R at 20 trades and ±0.34R at 100 for a typical 40% method.
• The sample-size ladder is a house policy that decides in advance when a method may be judged; it blocks both quitting on a bad run and trusting a lucky one.
• Score execution daily and the method only at review points, one variable at a time.

Further reading

• Outcome bias in decision evaluation (Baron & Hershey, 1988): the experiments showing people rate the same decision better when it happened to turn out well.
• Random Entry Trading (Jigsaw): a practitioner’s look at what small samples of entries can and cannot prove.
• Our lab: the streak and interval numbers in this lesson are exact arithmetic we checked; how long losing runs are in real NQ trade logs, and how much they cluster, is not yet measured.

Connects to: M7.06 (the emotions that break sample discipline) → M7.08 (the drawdown rules that keep the sample alive long enough to judge). RSK1.02 owns the streak maths; EV3.01 owns interval width and power.

Educational content only. Not financial advice. Futures trading involves substantial risk of loss.