Building a repeatable decision process
Turning a set of observations into a sequence that produces the same decision twice.
11 min read · intermediate · lesson 8 of 8
What this covers
- State why a fixed sequence produces a measurable method and an ad hoc one does not
- Write down the order in which observations are made
- Define entry and exit conditions before either is needed
- Record enough per decision that the method can be evaluated later
- Change the process on evidence rather than on the last outcome
The same inputs have to produce the same answer
Two people given identical charts should reach the same decision from the same process. So should the same person on two different days. Where that does not hold, there is no method to evaluate, only a series of individual judgments.
It fixes where judgment enters rather than removing it. A process specifies which observations are made, in what order, and what each one is allowed to decide. Judgment operates inside that, and its results become measurable because everything around it is constant.
One test settles it. Could the decision be reconstructed from notes a month later? If not, the outcome cannot be attributed to anything, and the record is a list of results rather than evidence about a method.
A sequence that holds
Condition
What state is the market in on the higher timeframe. Recorded before anything else, so it cannot be adjusted to fit what comes later.
Location
Where price sits relative to the levels that matter. Either it is at one or it is not; being near one is not being at one.
Evidence
What confirms the read. Named in advance, so a confirmation is recognised rather than constructed.
Invalidation
The level that would prove the read wrong. Located before size is considered, and never moved to accommodate a size.
Size
Risk amount divided by the invalidation distance. An output, not a choice.
Exit conditions
What ends the position other than the invalidation level: a target, a condition change, or elapsed time.
Unwritten processes are not processes
A sequence held in memory changes quietly under pressure. The step that was inconvenient last week is skipped this week, and there is no record that it was ever there.
Written down, a deviation is visible as a deviation. The written version is not better. It is checkable, and departures from it can be counted.
What each decision needs recorded
- The read
- The condition and location, in one line, as they were at the time. Written before the outcome is known, or it is a reconstruction.
- The evidence
- What was taken as confirmation. This is the field that later separates a method that works from a run of favourable outcomes.
- The invalidation level
- Where the read would be wrong, and whether that level was respected.
- The deviation
- Any step skipped or altered, named explicitly. Blank is a valid and useful entry.
- The outcome in R
- Result as a multiple of the risk taken, so entries are comparable across instruments and sizes.
Reviewing the process rather than the results
A losing decision that followed the process is not a mistake. A profitable decision that skipped three steps is. Sorting the record by whether the process was followed, rather than by outcome, is what makes review useful.
Four groups emerge: followed and profitable, followed and not, deviated and profitable, deviated and not. The third is the dangerous one. It rewards the deviation, and it is the group people remember.
What the record answers over enough entries is narrow but valuable: whether the deviations are helping, and whether the conditions the method claims to handle are the ones it actually handles.
Changing the process after the last outcome
The strongest pull to change a process arrives immediately after a loss. That is the moment the evidence is thinnest, and one outcome cannot distinguish a broken method from an ordinary losing run.
Set the review interval in advance: a fixed number of decisions, or a fixed period. Changes are made at the review, from the record, one at a time. Change two things at once and neither can be evaluated afterwards.
A process that changes after every loss is the same ad hoc judgment as before, wearing a written document.
One decision, recorded
IllustrativeAll figures are illustrative.
Condition: daily in an established uptrend, higher highs and higher lows intact.
Location: price returned to the prior structural high at 1.0850, now the nearest level below.
Evidence: the level held on the first test and the four-hour range contracted into it. Both were named in the process beforehand.
Invalidation: 1.0805, below the level and below the swing low that formed on the test.
Size: 1 percent of the account divided by 45 pips.
Exit conditions: prior high at 1.0990, or a daily close below the invalidation level, or ten sessions elapsed.
Deviation: none.
Outcome: closed at 1.0960 on a condition change, plus 2.4R.
The entry is useful a month later because the read, the evidence and the deviation field were all written before the result existed.
Key takeaways
- The same inputs must produce the same decision, or there is no method to measure.
- A process fixes where judgment enters rather than removing it.
- Invalidation is located before size, and size is an output of it.
- Sort the record by whether the process was followed, not by outcome.
- Deviated and profitable is the entry that does the most damage.
- Change the process at a scheduled review, one thing at a time.
Common mistakes
- Writing the read after the outcome is known.
- Moving the invalidation level so a preferred size fits.
- Judging the process by the last result rather than by the record.
- Changing several rules at once, leaving none of them evaluable.
Knowledge check
OptionalThe journal is where the read, the evidence and the deviation get recorded at the time rather than reconstructed later. Without that record the review described here has nothing to sort.
Related
- Confirmation and invalidationMarket Intelligence
- Multi-timeframe analysisMarket Intelligence