Part 4: Trading and the system · Chapter 13 of 14 · 10 min
A levels system on trial
Four variants built from what the chapters found (bounce at time-based levels, without round numbers, with confluence, break of H4 swings) give mean R from −0.05 to +0.03 (errors of ± 0.08-0.10 over 367-552 trades). The best is exceeded by 32% of trials with random levels; chosen on the first half, the break of H4 swings is worth −0.03 ± 0.14 in the second. To measure R with ± 0.05 about 1,500 trades would be needed, that is two to three years.
From measurements to a system
In chapters 1-12 we measured levels from many sides. A system puts them together: an entry, an exit, a filter and risk management, all written beforehand. Ours starts from four ideas the chapters suggested. It must be said at once that the four ideas come from the same data on which we test them: so the result on this sample is always favourable to the rule, and only the comparison with chance, the two halves and random levels counts.
- A: bounce at time-based levels. Previous day, previous week and Asia: entry at the close of the contact candle in the direction of the bounce, stop half an ATR beyond the zone, target 1 R, time exit after 24 candles, cost 0.3 $, one trade at a time.
- B: like A, only without a 50 $ round number within half an ATR. Idea from chapter 10: levels near a round number hold 44%, the others 51%.
- C: like A, only with at least two families of levels within half an ATR (confluence). The popular idea of chapter 10.
- D: break of H4 swings. Entry at the close of the first candle beyond the zone, stop a quarter of an ATR inside the zone, target 1 R. It is the rule with the highest R of chapter 11 (+0.03).
What happened
| Variant | Trades | Mean R ± error | Winners | Random direction | First / second half |
|---|---|---|---|---|---|
| A: bounce, time-based levels | 552 | −0.03 ± 0.08 | 49% | −0.04 | +0.06 / −0.12 |
| B: without nearby round numbers | 404 | −0.00 ± 0.10 | 51% | −0.05 | +0.04 / −0.03 |
| C: with confluence | 486 | −0.05 ± 0.09 | 48% | −0.04 | −0.00 / −0.10 |
| D: break of H4 swings | 367 | +0.03 ± 0.10 | 52% | −0.03 | +0.10 / −0.03 |
A note on the random-direction column: it is negative even though the sign is random. The reason is the cost (about 0.01 R) and the prudent rule by which a candle touching both stop and target counts as a stop. The right comparison for a trade is with that number, not with zero.
The readings:
- C, confluence, is the worst (−0.05 ± 0.09) and in the second half falls to −0.10: more nearby levels improve nothing (chapter 10).
- B reaches zero. Removing round numbers takes R from −0.03 to −0.00 and cancels the losses of entries at round numbers. But it stays zero, not positive, with an error of ± 0.10.
- D is the only one above zero (+0.03), with an error (± 0.10) larger than the result, +0.10 in the first half and −0.03 in the second.
Against random levels
For each variant we redid everything, filters included, on 25 series of levels shifted by 2-6 ATR:
| Variant | Real | Random, median (5th-95th) | Share of random trials with R at least equal to the real one |
|---|---|---|---|
| A | −0.03 | −0.06 (−0.12 / 0.00) | 32% |
| B | −0.00 | −0.05 (−0.09 / 0.00) | 4% |
| C | −0.05 | −0.05 (−0.13 / +0.04) | 52% |
| D | +0.03 | −0.01 (−0.06 / +0.06) | 20% |
B is the only one that stands out, barely: only one random trial in 25 reaches it. With random direction at −0.05, B has an advantage of 5 points: if true, it would be worth 0.05 R per trade, and with an error of ± 0.10 we do not know. Besides, we looked at four variants: the best of the four has R = +0.03 (D), and in 32% of random trials the best of the four is at least that high. Selection among variants explains the value.
The strictest check: we choose the best variant on the first half (D, +0.10 R) and test it on the second: it is worth −0.03 ± 0.14 R over 190 trades. The advantage seen in the first half disappears.
Read the chart in words
H1 gold chart of 20-21 January 2026 with the zone of the Asian high, the entry, stop and target of a variant B trade.
| When | What | Price |
|---|---|---|
| 20 Jan 2026 01:00 UTC | Asian level 4717.99 | 4,714.56 – 4,721.42 |
| 20 Jan 2026 13:00 UTC | entry | 4,739.05 |
| 20 Jan 2026 13:00 UTC | stop | 4,707.26 |
| 20 Jan 2026 13:00 UTC | target 1 R | 4,770.84 |
| 20 Jan 2026 23:00 UTC | target: +0.99 R | 4,778.92 |
Times in UTC.
Read the chart in words
H1 gold chart of 22-23 January 2026 with the zone of yesterday's close, the line of the round number 4,950 and the skipped contact.
| When | What | Price |
|---|---|---|
| 22 Jan 2026 12:00 UTC | yesterday's close 4938.66 | 4,932.69 – 4,944.63 |
| 22 Jan 2026 12:00 UTC | round number 4950 | 4,950.00 |
| 23 Jan 2026 00:00 UTC | contact skipped by the system | 4,944.53 |
| 23 Jan 2026 08:00 UTC | (it would have broken) | 4,914.98 |
Times in UTC.
B's curve
The cumulative R of B ends at −0.1 R over 404 trades (random direction at −20.9 R), with a maximum of +12.2 R and a minimum of −3.8 R. Month by month: December −0.5, January +1.7, February +4.4, March −1.4, April −2.7, May +7.7, June −5.4, July +2.4, August −5.9, September −0.2, October −0.2. The maximum drawdown is 17.1 R and the longest losing streak is 5. With a 1% risk per trade, 17 R is 17% of the account: a system with zero R hurts as much as a losing one.
Read the chart in words
Daily gold chart from December 2025 to October 2026 with, below, the cumulative R of variant B and of random direction.
How long it would take
The standard deviation of each variant's R is about 1.0. To measure R with ± 0.10 about 380 trades are needed, for ± 0.05 about 1,520. The variants make between 38 (D) and 57 (A) trades a month: for ± 0.05 you need 27 to 41 months, that is from two and a quarter years to three and a half. More filters, fewer trades, more time to know.
A six-month forward test would give 230-340 trades: enough for an error between ± 0.11 and ± 0.13, not to say whether B's +0.05 exists.
Read the chart in words
H1 gold chart of 8-10 January 2026 with the zone of yesterday's close, the entry, stop, target and four candles marked with the letters A, B, C and D.
| When | What | Price |
|---|---|---|
| 8 Jan 2026 14:00 UTC | yesterday's close 4474.97 | 4,471.14 – 4,478.81 |
| 9 Jan 2026 02:00 UTC | entry | 4,467.80 |
| 9 Jan 2026 02:00 UTC | stop | 4,486.60 |
| 9 Jan 2026 02:00 UTC | target 1 R | 4,449.00 |
| 8 Jan 2026 22:00 UTC | Point A | 4,475.91 |
| 9 Jan 2026 02:00 UTC | Point B | 4,477.98 |
| 9 Jan 2026 12:00 UTC | Point C | 4,477.00 |
| 9 Jan 2026 13:00 UTC | Point D | 4,487.41 |
Times in UTC.
What to do with it
- Do not trade a rule with this error with real money. B gives −0.00 ± 0.10: the true value could lie between −0.10 and +0.10 R.
- Use it as a laboratory. The rules are written, the data are public, the log is a spreadsheet. Measure it on new data without risking anything.
- Fix the judgement criterion beforehand. For example: "after 400 trades, if R ± error still contains zero and random direction, I discard the rule".
- Compute the expected drawdown. With an R near zero, a 17 R drawdown is normal and means 17% of the account at 1% risk: decide beforehand how much you are willing to lose.
- Remember how the rule was born. The variants come from the same data: the real test is on data that did not inspire the rule.
Procedure in five steps
- Write the rule, the filters and the judgement criterion in advance.
- Test it on all available data, counting the variants.
- Compare with random direction, the two halves and random levels; correct for the variants tried.
- Compute the maximum drawdown and how long it would take to know.
- If it is not proven, measure it forward without real money, with a log, until the criterion gives an answer.
Exercises
Chapter exercises
1. On the chart the lines are the entry (4,467.80 $), the stop (4,486.60 $) and the target (4,449.00 $) of a variant A trade on 9 January 2026 (UTC hours). Which letter is the entry candle, that is the close of the contact candle?
Read the chart in words
| When | What | Price |
|---|---|---|
| 8 Jan 2026 14:00 UTC | yesterday's close 4474.97 | 4,471.14 – 4,478.81 |
| 9 Jan 2026 02:00 UTC | entry | 4,467.80 |
| 9 Jan 2026 02:00 UTC | stop | 4,486.60 |
| 9 Jan 2026 02:00 UTC | target 1 R | 4,449.00 |
| 8 Jan 2026 22:00 UTC | Point A | 4,475.91 |
| 9 Jan 2026 02:00 UTC | Point B | 4,477.98 |
| 9 Jan 2026 12:00 UTC | Point C | 4,477.00 |
| 9 Jan 2026 13:00 UTC | Point D | 4,487.41 |
Times in UTC.
2. Four variants give R from −0.05 to +0.03 and the best (+0.03) is exceeded by 32% of trials with random levels. What do you take from it?
3. System B has a maximum drawdown of 17.1 R with a mean R near zero. What does it mean with a 1% risk per trade?
4. The standard deviation of R is about 1.0 and the system makes 40-57 trades a month. How long to measure R with ± 0.05?
Have a question or want to share your exercise?
Post in the community, or join the free signals room on Telegram.
Educational content, not financial advice. Trading involves risk.