"I make money" is not information: it does not say with what. Tagging every trade with its setup turns two hundred rows into four or five questions that do have answers: which of your ideas pays, which costs you money and which does not yet have enough sample to know. Here is how to build that list of setups, how many trades it takes and what to do with the resulting table.
Setup, context tag and mistake tag
It pays to separate three things that usually get mixed into one field:
- Setup: the idea that justifies the entry. Opening breakout, pullback to VWAP, failed high. Every trade has exactly one, decided before entering.
- Context tags: conditions you want to slice by later: session (open, midday), news day, day's trend, instrument. A trade can have several.
- Mistake tags: what you did wrong, if anything: early entry, moved stop, size off plan, trade outside hours. Applied in the review, not in the heat of the moment.
The separation matters because they answer different questions. The setup answers "which ideas work?"; the context, "when do they work?"; the mistake, "how much does not following the plan cost me?". That last number is usually the largest of the three.
A taxonomy that does not break
The enemy of a setup list is the list that grows every week. What works is a closed list of three to eight setups, each with a stable name and a one-sentence definition that says when it does not apply. If you need to distinguish variants, do it with context tags, not new setups:
| Setup (one per trade) | One-sentence definition | Context (several) |
|---|---|---|
| Opening breakout | Break of the first 15-minute range with volume; does not apply on FOMC days. | open · trend · news |
| Pullback to VWAP | Entry on the first touch of VWAP with the day's trend; does not apply if price crossed VWAP twice. | midday · trend |
| Failed high | Reversal after a high that does not exceed the previous one with a volume divergence; range days only. | range · close |
Write the definitions in the journal. In three months you will not remember whether "pullback" included the second touch, and if the definition changes, the earlier statistics no longer hold.
How many trades you need
With ten trades, a setup with a 70% win rate has proven nothing. The standard deviation of the observed win rate is roughly the square root of p × (1 − p) ÷ n: with p = 0.5 and n = 10 that is 16 percentage points, so an observed 70% is consistent with a real 50%. With n = 50 it drops to 7 points; with n = 100, to 5. The practical rules:
- Under 20 trades: no verdict. Keep logging, at small size.
- 30 to 50: you start to see the sign of the expectancy and the big differences between setups.
- 100 or more: you can fine-tune details like stops, targets and time windows.
Expectancy in R
Expectancy is what you expect to make per trade. In R, with the initial risk as the unit, it is computed like this:
expectancy (R) = win rate × average win (R) − loss rate × average loss (R)
Example: 40 trades of one setup, 22 winners averaging +1.8R and 18 losers averaging −1.1R (some slipped past the stop). Win rate 55%, loss rate 45%: 0.55 × 1.8 − 0.45 × 1.1 = 0.99 − 0.495 = +0.495R per trade. With a $100 R that is $49.50 gross per trade; with $4 of round-turn commissions, $45.50 net. Measuring in R rather than dollars is what lets you compare this setup with another traded in a different instrument or at a different size, as the guide on position sizing explains.
The per-setup table and the three decisions
Every week, a table like this one, and for every row one of three decisions: keep, fix or kill.
| Setup | n | Win rate | Avg win | Avg loss | Expectancy | Decision |
|---|---|---|---|---|---|---|
| Opening breakout | 64 | 48% | +2.1R | −1.0R | +0.49R | Keep: it carries the account |
| Pullback to VWAP | 41 | 61% | +1.1R | −1.3R | +0.16R | Fix: losers exceed 1R, the stop is being moved |
| Failed high | 37 | 38% | +1.4R | −1.0R | −0.09R | Kill, or cut to micros until n = 60 |
| Midday impulse | 9 | 67% | +0.9R | −1.0R | +0.27R | No verdict: continue at minimum size |
- Keep: positive expectancy with enough sample. That is where size goes.
- Fix: positive expectancy but with a clear symptom, such as average losses above 1R (the stop is being moved) or wins below the target (exits out of fear). One variable is corrected and measured again.
- Kill: negative expectancy with enough sample. Out of the plan. If it hurts, test it with micros until the sample doubles, not at the usual size.
Overfitting: when every trade is unique
The opposite mistake to not tagging is tagging too much. With twenty setups and fifteen context tags you will always find a combination with 80% accuracy over five trades, and it will always be noise. Three defenses:
- A closed list of setups; variants go into the context.
- Never cross more than two dimensions at once (setup × session, not setup × session × instrument × weekday).
- No slice with fewer than 30 trades changes a decision.
If a slice looks revealing, write it down as a dated hypothesis and wait for the sample to confirm or kill it. Half the time it kills it.
The same trade in several accounts
If you copy trades across several prop firm accounts, every trade appears as many times as you have accounts. Counting them separately inflates the sample without adding information: five accounts with the same entry are one decision, not five. For setup statistics, count each position once; for risk management, look at each account separately, because each has its own cushion and daily limit.
The weekly routine
- Every day at the close: tag each trade's setup (if you did not at entry) and the mistake tags. Five minutes.
- On Friday: the per-setup table with n, win rate, averages and expectancy. Cross with session only if a setup has more than 30 trades.
- One written decision, dated, with the sample behind it. One, not five.
- Every month: review the setup definitions. If one changed, mark the date and start a new sample.
Frequently asked questions
How many setups should I have?
Between three and eight. With fewer than three you cannot tell anything apart; with more than eight none of them reaches the 30 trades you need to judge it within a reasonable time. If you have twelve, group them: they are almost always variants of four ideas.
Do I tag before or after the trade?
The setup, before: it is the reason you enter and cannot depend on the result. Mistake tags, after, in the daily review. If you tag the setup after the fact, you will tend to call "breakout" whatever worked and "impulse" whatever did not.
What if a setup has positive expectancy but a 35% win rate?
That is perfectly normal and usually a trend setup with distant targets: it loses often and small, and wins rarely and big. What you must watch in those setups are the streaks: at 35% accuracy, eight losses in a row happen frequently, and your risk plan has to survive them.
Next step
Write your closed list of setups today, each with its one-sentence definition, and tag the trades of the last two weeks. If you do not yet record every trade's initial risk, start with what to log in a futures journal; without that number there is no R, and without R there is no comparable expectancy. And so the streaks of low-win-rate setups do not take the account, size with the risk ladder.