Checklists over willpower – how to design rules that hold in the worse moment

A plan is usually written in calmer conditions than those in which it must be executed. During a session, time pressure, an open position and a recent outcome enter the picture—so “I will be disciplined” may be insufficient. This article shows how to turn an intention into a condition, a specific action and a record that makes execution checkable. We consider checklists, restricting choices in advance, and the cost of excessively rigid procedures. We also separate research findings from practical proposals for traders. Article 10.2 covers state and impulse; article 10.5 covers accumulating violations after a loss.

Preparation theme: define the condition, action and audit trail before the order. The illustration does not prove that a particular checklist was completed.

A checkmark needs a value, a time and a version

A checklist can be complete and still describe conditions that no longer exist. “Spread checked” says little unless the record identifies which quote was checked and what happens when it changes. A useful procedure makes that dependency explicit rather than asking the reader to remember whether the market has moved since the box was marked.

For a hypothetical replay, use a long EUR/USD candidate of 40,000 EUR on a USD account. A pip is 0.0001 USD per EUR, worth 4 USD at this volume. The procedure requires spread no greater than 1.2 pips, scenario loss no greater than 55 USD and a quote update no older than 500 milliseconds at assessment. These are invented exercise limits, not recommended trading parameters. All shown times use one synchronised model clock in UTC; a real record must establish its own clock and timestamp meanings.

Attach the checks to the quote that supplied the inputs

Two hypothetical updates for the same candidate
Quote versionUpdate time UTCBidAsk
Q1712:00:00.0001.249001.24910
Q1812:00:00.4001.249001.24918

At 12:00:00.100, the record checks Q17. Its update is 100 milliseconds old and its spread is 1 pip. Assume a protective exit filled at Bid 1.24790 and a total round-trip commission of 4 USD, with no other charges or financing. If the entry fills at Q17’s Ask, scenario loss is 12×4+4=52 USD. All three stated tests pass for those inputs.

The hypothetical rule also says that a changed price input invalidates the dependent spread and loss checks before an order is requested. The record stores the quote version alongside the exit level, volume and tariff. MetaQuotes’s price-data structure exposes Bid, Ask and an update time in milliseconds. That provides fields to record; it does not establish clock synchronisation or a universally acceptable age threshold.

Recheck the latest version instead of carrying forward the old boxes

At 12:00:00.450, Q18 is the latest update. It is only 50 milliseconds old, so the model’s age test passes. Its spread is now (1.24918−1.24900)/0.0001=1.8 pips, above the limit. If entry instead fills at its Ask with the same stipulated protective exit, the distance is 12.8 pips and scenario loss is 12.8×4+4=55.20 USD. That also exceeds the loss ceiling, by 0.20 USD.

The earlier 52 USD calculation is not a new calculation for Q18. The quote has become more recent while two other conditions have become unacceptable. Recording “all checked” at the later moment without updating the dependent fields hides that difference. Under this exercise procedure, the candidate fails the current gate. This says nothing about whether the next price movement will be favourable.

Design the failure response before the check becomes inconvenient

Write down what invalidates a check, which fields must be repeated and what confirms that the required action occurred. In this example, the response is to stop the proposed submission and reassess using a consistent set of current inputs. It is not to change the limits after seeing a failed value, nor to retain the old risk number while attaching a newer timestamp.

This is a proposed record design, not a demonstrated psychological intervention or a guarantee against errors. A checklist can enforce a badly chosen rule perfectly; an input can also be missing or incorrect. Keep those cases distinguishable from a rule deliberately bypassed. Review revisions outside the decision being assessed, and preserve the earlier version so the review does not rewrite its standard. The detailed lesson examines if–then plans, precommitment, research limitations and the cost of excessive rigidity. Here the first practical question is narrower: do the marked checks still refer to the facts used for the action?

Three exercises with worked answers

The exercise data are hypothetical. Choose an answer, then check the calculation.

1. How old is Q18 at the later assessment?

Q18 update 12:00:00.400 UTC; assessment 12:00:00.450 UTC, using the same model clock.

Show the answer and explanation

C. Subtract timestamps from the same clock and retain the meaning of the measured interval.

  • A. That confuses the assessment’s millisecond field with elapsed time.
  • B. That is the update’s millisecond field, not the difference.
  • C. Correct: the two stated times differ by 50 milliseconds.
  • D. That is the allowed maximum, not the measured age.
2. What scenario loss follows from the Q18 Ask under the stated assumptions?

40,000 EUR; entry 1.24918; protective fill 1.24790; pip 0.0001, worth 4 USD; commission 4 USD.

Show the answer and explanation

A. Recompute every price-dependent field for the input version actually being assessed.

  • A. Correct: 12.8×4+4=55.20 USD.
  • B. That carries forward Q17’s price input.
  • C. That omits the separate commission.
  • D. That substitutes the ceiling for the calculated amount.
3. What should the procedure conclude at the later assessment?

The age test passes, but the latest spread and loss exceed their unchanged limits.

Show the answer and explanation

D. Bind each conclusion to its inputs, and state the response to failed or missing values in advance.

  • A. Those marks described earlier inputs.
  • B. Recency does not establish an acceptable spread or loss.
  • C. That changes the rule during the decision being judged.
  • D. Correct: passing one current check does not preserve the other expired checks.

Continue to procedure design and its limits

The detailed lesson separates resource theories of self-control from the evidence, explains if–then formats and precommitment, and examines checklists that become ineffective or too rigid. It keeps findings from other domains separate from proposals for trading practice.

1. Why a plan written calmly fails under pressure

You may prepare a rule with no open position or time pressure, then execute it after a loss or in a hurry. The conditions are not always opposite, but the difference matters for design. A state change may make your own action harder to predict accurately. It does not establish a change in character or that every calm forecast will be too optimistic. Check both the plan’s content and the conditions in which it is to be executed.

The phenomenon has a name and a literature. George Loewenstein described it as the cold–hot empathy gap: in a cold state people underestimate how strongly a hot state – hunger, pain, fear, arousal, time pressure – will change their choices, and in a hot state they underestimate how quickly that state will pass[1]. The original studies in this strand were run in laboratories, on student samples and with non-financial stimuli. What transfers to trading is the mechanism alone, and that is how it is used here; the effect size stays with the population it was measured in.

The design consequence concerns the same person in different conditions. If pressure permits a rule to be changed during execution, thinking it through beforehand may not be enough. A plan can restrict reversibility or specify when and on what evidence it may be updated. Another decision in a difficult moment is not always an error; the problem is an uncontrolled opening through which the previously adopted condition can be bypassed.

What the cold–hot gap does not say. It does not say that decisions made in arousal are always worse. Nor does it say that emotion is an error – a fast reaction is sometimes the correct reaction. It says something narrower and more useful: a forecast of your own behaviour, made in a different state from the one the behaviour will occur in, is biased in a predictable direction. That is why the object of design is the rule; attitude does not submit to design.

A call for more discipline does not specify an action or condition. Repeated violations warrant checking the rule’s design, feasibility, enforcement and circumstances. They do not establish in advance that the problem lies exclusively in design or character. Specify a concrete change that can later be compared in the records.

2. What actually happened to "willpower as a resource"

The popular justification for procedures sounds different from the above: self-control is supposed to be a resource that gets used up, so towards the end of a session "you run out of fuel". That claim has a specific origin and a specific status, and both should be stated rather than the slogan repeated.

The model was proposed by Roy Baumeister and colleagues in 1998 on the basis of a series of laboratory experiments: performing a task requiring self-control worsened persistence on a subsequent, unrelated task[4]. The participants were students, the stakes were symbolic, there were no financial consequences. In 2016 the effect was tested in a preregistered replication report: twenty-three laboratories, a common protocol agreed before data collection, 2,141 participants in total. The result: d ≈ 0.04, with a confidence interval covering zero – an effect indistinguishable from zero[5].

The agreed protocol did not demonstrate an effect distinguishable from zero. That does not establish that no effect exists in every task, or end the debate about self-control models. A later preregistered test across 36 laboratories and 3,531 participants likewise found a small, non-significant effect: d ≈ 0.06[15]. These findings do not justify presenting “willpower fuel depletion” as an established law from which specific trading recommendations follow. Procedure design does not require adopting that model.

Dang et al. (2025) add to this picture: across 14 samples (N = 2078), a 30–40-minute antisaccade task requiring inhibition of reflexive gaze was followed by more errors on No-Go trials than in controls (d = 0.31–0.35, depending on the comparison group). The initial experiment and direct replications were preregistered; later extensions were not. This is a result from a specific protocol, rather than proof of “fuel” depletion or a trading test[16].

Implications for procedure design. The resource metaphor is not enough to recommend food, a break or training as a way to restore self-control in trading. Rest may have other justifications, but those need their own evidence and assessment of conditions. The cold–hot gap draws attention to a difference between prediction and execution. An earlier condition and constraints on bypassing it are practical design proposals from that perspective, rather than a guarantee that pressure will leave no further decision. A design can be assessed without resolving the entire resource-model debate.

The same applies to the most frequently cited "evidence" for decision fatigue. A 2011 study of Israeli parole board decisions showed that the share of favourable rulings fell almost to zero over the course of a hearing block and returned to a high level after a break[6]. These are observational data without random assignment, however, and a critique published shortly afterwards pointed out that the order of cases was not random: cases with legal representation were heard earlier, and cases from the same prison came in blocks[7]. The alternative explanation – simply a different mix of cases in different parts of the block – has not been ruled out. The "hungry judge" is not an established fact and will not serve as a foundation for anything in trading.

What is left of this for a trader. An observation, not a law: execution quality is sometimes worse towards the end of a session, after a run of losses and with a larger number of decisions. That is a hypothesis you can check in your own record – by comparing the share of orders placed without the full set of required steps in the first and second halves of the session, across a run of several dozen sessions. How to keep such a record is described in article 10.8. Until you have that data, "I am worse at the end of a session" remains a supposition.

3. If–then plans – the format, the evidence and the limits of that evidence

One technique with quantitatively summarised evidence is if–then planning, or implementation intentions. Peter Gollwitzer developed the concept[2]. Gollwitzer and Sheeran’s 2006 meta-analysis covered 94 independent tests and more than 8,000 participants, estimating d ≈ 0.65 for attaining previously adopted goals[3]. This is a medium-to-large effect in the tasks studied, rather than a financial performance improvement or evidence that this is the only effective technique.

The limits of that evidence have to be stated alongside it. The studies mostly concerned health, educational and consumer goals – regular exercise, screening tests, handing work in on time. The populations were mainly student and general, the measurement horizon short, and the effect larger where the goal had already been adopted and execution alone was the problem. This meta-analysis does not establish effectiveness in retail FX traders. So what transfers to trading is the format and the mechanism – moving the decision onto a previously fixed, observable condition – and not the number 0.65.

The format links an event to a specified action: if [observable condition], then [specific action]. The condition should have an unambiguous checking method, and the action a defined execution order. Some conditions require a reading or calculation; prepare the data or an automatic signal before pressure arises. "I will be patient" does not state when to act or what to do. The examples below show the form of a rule, rather than universally appropriate behaviour.

Resolution (unsettleable)The same intention in if–then format
"I will not increase size after a loss"If the volume in the order window differs from the one recorded before the session, I close the window without sending the order
"I will only trade my own setups"If the instrument is not on the list recorded before the session, I do not open its chart
"I will be patient"If the clock shows the closing time of my window, I log out regardless of what I see on the chart
"I will not move my stop loss"If the protective order is already in the market, the only permitted change is the one recorded before entry; any other means closing the position
"I will not trade when I am rattled"If there is no completed pre-session record with a time of completion, the session runs in observation mode
"I will take a break when things go badly"If a second procedure violation appears in the session record, new orders are blocked until the review is finished
"I will log my trades"If the rationale is not written down, the order is not sent – the order of operations is part of the rule

The right-hand column is not better because it sounds sterner. It is better because someone else can check each of these conditions in your record and reach the same conclusion you did. The left-hand column has none of that, and so mood settles it.

When a condition makes execution harder. A lengthy data review or undefined interpretation may delay a response under pressure. That does not mean every condition requiring calculation is ineffective. Define the method beforehand, prepare the reading and specify the permitted response time. A clock time, the presence of an entry or an order’s status are simple examples. A review flag: comparable records do not establish whether the condition occurred, or it is recognised only after the action it was intended to stop.

4. The four fields of a procedure and the failure criterion

"Be disciplined" is not a procedure, but not because it is trite. It is not a procedure because it lacks all four fields that must be present for a rule to be executable and checkable. That set of four is the standard form across the whole of section 10: later articles apply it and refer back here instead of redefining it.

Trigger condition. The event after which the procedure starts. Its recognition method should be defined beforehand and checkable. "When I feel I am losing control" is a self-reported condition, but without level descriptions and a response rule it leaves considerable room for interpretation. It can be supplemented by a history event or defined scale; that does not turn a self-rating into an objective test.

Action. One, specific, executable without deliberation. Not "think the situation over" but "close the order window", "log out", "record the event in the file". If there are several actions and you have to choose, the procedure hands the decision back to the place it was taking it from.

Completion criterion. How you know the procedure has ended and you are returning to normal mode. A rule without an exit condition is not a rule, just a mood of longer duration. "Come back when you have calmed down" has no completion criterion – it is judged by the same state that triggered the procedure.

Audit trail. Information that can test execution: a time, condition value, block status or number of subsequent orders. It may come from a form or system history. Without an appropriate trace, some questions remain unresolved; a missing manual note does not establish that an action was omitted if another reliable record confirms it. The trace should match what the procedure was intended to do.

Beside the form stands a requirement that does not belong to it, although the two are often confused. The failure criterion is the observation after which you are to accept that this particular rule is not doing what it was created for. It is not a fifth field of the procedure, because it does not describe execution – it settles whether the rule continues to exist. That is why section 10 requires one of every article and of every rule described, including in this text. Without it, a dead procedure looks exactly like a working one.

RuleTrigger conditionActionCompletion criterionAudit trailVerdict
"Be disciplined"nonenonenonenoneNot a rule, a wish
"Do not trade when emotions are above neutral"not observabletwo to choose fromnonenoneSettled by the state it was meant to detect
"Do not touch the position unless conditions change fundamentally"open-ended categorynonenonenoneAn escape hatch wider than the rule
"Take a break until you calm down"not observableoneunsettleablenoneNo exit checkable from outside
"If a second violation appears in the session record, new orders are blocked until the review is finished"observableonereview completedviolation code and time of the blockComplete – this is a procedure

The last row requires a caveat that must not be skipped: I am not claiming that the "second violation" threshold is the right one. There is no data that settles it, and any threshold given here as universal would be invented. What is exemplary here is the form of the rule; the parameter is set separately. The parameter follows from your own record and from the risk management plan, which is the subject of section 9.

The last row of the table written out in four fields.
Trigger condition: a second violation from the code list appears in the current session's record. The condition is read from the file and a third party will settle it the same way.
Action: new orders are blocked immediately after the second code is recorded, before the next look at the chart.
Completion criterion: the review of the current session is done and its result takes the form of one sentence in the journal. The block ends at the moment that sentence is written and at no other.
Audit trail: the time the second code was recorded, the time the review closed and the number of orders created between one and the other. Three values, all readable from the export.
Failure criterion (a requirement separate from the form): new orders appear despite the active block, or the block is bypassed. No triggers across several dozen sessions call for checking whether the condition had an opportunity to occur; absence alone does not establish ineffectiveness. A trigger in every session calls for reviewing working conditions and the definition rather than automatically relaxing the rule.

The four fields and the assessment requirement attached to them recur throughout this section wherever the word "procedure" appears. That is how the countermeasures in article 10.2 are built, along with the recording procedures in article 10.8, the three phases of the reset protocol in article 10.9 and the rule limiting trading under reduced capacity in article 10.11. There is one form and this text defines it; those articles fill it with their own content.

5. A rule settleable without judging your own state

This is the point at which most sets of trading rules break in a way that is hard to notice. The rule sounds sensible, looks like a procedure, even has an action – and yet it safeguards nothing, because what decides whether it fires is exactly the state it was meant to catch.

The classic example looks like this: if emotions are above neutral, trade smaller or do not trade at all. Three faults in one sentence. The condition is not observable, so nobody but you will check it. It is judged by the person whose judgement is precisely what is in doubt. And it ends in an alternative, which hands the decision back into the moment it was supposed to be taken out of.

Rewriting it means swapping a state for an event. An event has the property of either being there or not, and someone else can establish that by looking at the record:

  • Procedural condition. If the full set of pre-session fields, along with the time of completion, has not been filled in and saved before the session, the session runs in observation mode. Completion criterion: the close of the session by the clock – observation mode does not carry over to the next day. Audit trail: the time the pre-session record was completed, set against the time of the first order. There is no room here for "I am probably fine today".
  • Event condition from a closed list. You write the list of events that exclude trading once, calmly, and enter the type of event into it; felt intensity stays off the list. A sleepless night on call, illness at home, a job interview scheduled for that day. During the session you check only whether an event from the list occurred – you do not judge how much it affected you. Completion criterion: the end of the day on which the listed event occurred. Audit trail: the date and the item number from the list, entered in the journal before the session.
  • Condition from order history. If a procedure violation is already recorded in the current session, the next order requires a review to be carried out. You read the condition from the file. Completion criterion: the review recorded in the journal. Audit trail: the violation code and the time the review closed.

Failure criteria for this family concern a recognised event that the procedure failed to stop, execution only after an order, or removal of the barrier during the session. No triggers across several dozen sessions is a reason to check whether the condition had an opportunity to occur—not proof that the rule is dead. Frequent triggers call for examining working conditions and the definition, rather than automatically relaxing it. Record a change after review, outside the session.

Self-report is data about a declared state, rather than an objective reading of it. A “how do I feel today, 1–10” scale can compare personal records and, under a rule defined beforehand, may also trigger a procedure. Its limitations include dependence on judgement and the possibility of adjusting an answer. Keep the question, level descriptions and measurement timing fixed, and supplement the declaration with a checkable event where possible. The scale alone neither detects tilt nor establishes the cause of a violation.

6. Hard and soft precommitment and three levels of safeguard

Precommitment is a deliberate, earlier restriction of one’s future options. The cost or delay of reversal is an important safeguard feature. It does not by itself explain effectiveness: the block’s scope, enforcement and available bypasses also matter. A declaration in a file may support remembering a rule, but should not be equated with technically preventing an order.

That people do use this option – and that they set it up sub-optimally – is shown by Ariely and Wertenbroch's 2002 experiment. Students had to submit three papers over a semester. One group was given interim deadlines imposed at even intervals, a second could impose deadlines on themselves with a penalty for overrunning, a third had only a single final deadline. The self-imposing group did better than the group with one deadline, but worse than the group with imposed deadlines[11]. The conclusion has two halves and both matter: restricting your own options works, and at the same time we set those restrictions worse than an outsider would. The study concerned students and coursework – what transfers to trading is the mechanism alone.

Soft and hard – the operational difference

Soft precommitment is a declaration: a note, a promise made to someone, an alert, a rule written in a file. Reversal is easy, although a promise made to another person may carry a social cost. Hard precommitment is a commitment whose reversal requires time, another person or action outside the platform.

Before the session, check who can remove the safeguard, how and within what time. If it can be disabled immediately, it may be easily reversible even when technically advanced. Comparing that time with order placement is not a validated threshold. Client-side or server-side location alone does not settle the issue either: actual permissions and an operational test are needed. These differences are described in article 10.5; risk parameters remain outside this comparison.

Three levels of safeguard

Three levels of safeguard — ordered by the cost of removal 1. Technical A setting in the platform, at the broker or in an external tool. Scope and configuration belong to the platforms section, not to psychology. removal: seconds to several days — depending on the side 2. Procedural A rule requiring an action that leaves a trace in the record. You can lift it yourself, but lifting it also leaves a trace and can be counted. removal: immediate, but visible in the data 3. Physical A change of surroundings: no app on the phone, capital held separately, hardware out of reach, another person owning the change. removal: at the cost of time or someone else's consent The levels do not replace one another. A client-side technical safeguard can be weaker than a physical one.
The level of a safeguard does not depend on how advanced the tool is, only on how much it costs to remove mid-session. An ordering schematic, with no numerical values.

The split looks obvious only once the criterion has been introduced. The criterion is: who can remove the safeguard, and in what time. A programmable add-on on your own computer is technically advanced and at the same time soft, because it is switched off with one click. Separated capital is primitive and harder, because it works through the time a transfer requires.

Seven barriers described by form, without numbers

The set below is material moved over from the article on account destruction, where it sat in the wrong place. None of these barriers contains a threshold or a parameter – numerical thresholds belong to section 9 and that is where they are to come from. Here the subject is the shape of the safeguard and nothing else. Each barrier is given its failure criterion: the observation from the record after which the barrier has to be rebuilt.

  1. A limit enforced on the server side. The threshold comes from your risk plan; what matters is that you do not decide on its removal alone and not during the session. Failure criterion: you can point to a session in which the limit was raised on the same day.
  2. A protective order submitted together with the entry, not after it. The order of operations is the entire content of the rule here. The reading that invalidates it: in the history export you find positions where the protective order's submission time is later than the entry time.
  3. A pending order instead of a market entry. The decision is taken when the condition is being described, before price touches it. The sign that the barrier is illusory: the share of market entries submitted outside the planned window is rising.
  4. A time window closed by the clock. The end of the session is set by a clock time; the feeling that "something is still happening" does not count here. How you will know it has stopped working: the login log shows returns to the platform after the window closed.
  5. A closed list of instruments. A chart off the list is not opened, so there is nothing to assess. The signal that it is not working: instruments appear in the history that were not on the morning list.
  6. A procedural gap between the condition and the order. The gap has to be filled with an action that leaves a record; "count to ten" leaves nothing and cannot be checked afterwards. The observation that refutes it: the record is created only after the order has been sent.
  7. A record before the decision as a necessary condition of the order. A barrier and audit material in one. When to declare it dead: rationales are entered in bulk after the session.
What this set does not guarantee. None of the seven barriers improves signal quality, raises hit rate or creates an edge. All of them do one thing: they reduce the number of decisions taken on the spur of the moment. If the rule set you attach them to has no positive justification, the barriers will merely tidy up the way you lose money. Assessing whether that justification exists at all – section 8.

Three barriers that are not in the platform

The hardest safeguards are often the least technical, because they operate outside the device on which the decision is taken.

The app deleted from the phone. It changes nothing about what you feel – it lengthens the path from impulse to order, and that is all it does. That is enough, because most spur-of-the-moment decisions rely on the path being short. Failure criterion: the export shows orders placed outside the planned window, from another device.

Capital split between the trading account and the bank account. The only active part of this mechanism is the time a transfer requires. A top-up made during the session removes it entirely and, incidentally, raises exposure exactly when the process has already changed. Proof that the barrier is gone: top-ups made on the same day the loss occurred.

Another person owning the change. Changing the limit requires their involvement, so it cannot be done quietly between one order and the next. How you will know it has stopped working: changes made without their knowledge, or their standing, blanket consent to everything.

When this stops being a question of procedure. If working around your own blocks keeps recurring despite successive attempts to hold them, if trading starts being funded with debt or its scale is being hidden from those close to you – the problem has stopped being a problem of rule construction and another barrier will not solve it. The right address then is a doctor or a psychotherapist. This article diagnoses nobody and judges nobody; it only marks the limit of its own competence.

The cost of rigidity

Precommitment has a price and it is fair to state it. A rule that is too rigid blocks the adaptation that happened to be rational. A ban on any intervention in an open position works as long as execution conditions are normal – and stops working when quotes disappear, the instrument is suspended or the spread widens many times over.

The answer, however, is not an escape hatch of the "unless conditions change fundamentally" variety, because such a hatch is wider than the rule and voids it in every situation in which you happen to want it voided. The answer is a closed list of measurable conditions, written out in advance – and the fact that the list is closed matters more than what ends up on it.

Protective order: planned placement and later changes. Conditions and inferential limits are part of the diagram.
Not every change is a mistake. Compare it with the prior rule. An SL is an order, not a price guarantee.
A submitted protective order is not yet an executed protective order. A barrier on your own platform's side is not responsible for what the market does. A protective order is an exit order, not a price guarantee: with thin liquidity, a price gap or slippage, the fill can land considerably further away than the level shown on screen. That is not an argument against the barrier – it is the limit of what the barrier protects against.

There is also something the barrier does not change, and it is worth saying plainly. The tendency to close winning positions too early and hold losing ones is empirically documented: Odean showed it across roughly ten thousand brokerage accounts of US equity investors between 1987 and 1993, while the concept of the disposition effect itself was introduced earlier by Shefrin and Statman[13]. Carrying that result over to the leveraged retail currency market is an extrapolation. Exit orders submitted in advance do not remove the tendency – they only take away the ability to act on it in real time, and that is their entire role. The mechanism itself is described in article 10.6.

7. Fewer decisions instead of better decisions

A checklist and decision architecture solve two different problems, and they get confused because both end in the word "procedure". A checklist improves a single decision. Architecture reduces the number of them. The second works regardless of whether the first is being executed, which is why it is worth designing separately.

  • A closed list of instruments. You are not assessing twenty charts worse – you are assessing three, because the rest are not open. The decision about what is on the list is taken once, outside the session.
  • Time windows. You set the hours in which you look at the market at all in advance. Outside the window there are no weak decisions, because there are none at all.
  • Pending orders instead of market entries. You describe the condition calmly, execution is automatic. What changes is not the quality of the description, only the moment at which it is created.

The mechanism here differs from the one in the if–then plan and has no comparable evidence behind it. This is workshop practice: no study measures how much reducing the number of decisions improves retail traders' results, and I am not pretending one exists. The justification is logical and only that – a decision that does not exist cannot be taken badly.

Failure criterion: if the number of orders per session does not fall after you narrow the instrument list, the architecture has displaced activity instead of reducing it. Three instruments watched at the same frequency as twenty give the same number of opportunities to intervene, only in a narrower place.

8. Checklists – what transfers from medicine and aviation, and what does not

The argument from surgery and aviation appears in every text about trading discipline and almost always appears without its label. The label matters here, because it changes what you are allowed to do with the conclusion.

The 2009 study by Haynes and colleagues covered the implementation of the nineteen-point WHO perioperative checklist in eight hospitals in eight cities of very different wealth levels; around 3,700 patients before implementation were compared with around 4,000 after it[8]. Perioperative mortality fell from 1.5% to 0.8%, and complications from 11.0% to 7.0%. In absolute terms that is a fall of 0.7 percentage points and of 4.0 percentage points; the same magnitudes given as relative reductions produce the frequently quoted 47% and 36%.

The limitations matter as much as the result. This was a before-and-after comparison in the same centres, without randomisation and without a control group; organisational changes were happening in parallel, and participation in the study itself may have changed team behaviour. Later large-scale mandatory rollouts did not reproduce the fall in mortality. What transfers to retail trading is the mechanism – under pressure, people skip steps they know by heart – and the mechanism only. The effect size does not transfer at all: a different population, different stakes, a team instead of one person, and external oversight that a retail trader does not have.

Three distinct settings for procedures: aviation, surgery and individual trading.
An analogy of format, not effectiveness. Haynes et al. (2009) studied checklist implementation in eight hospitals with a before–after design and no randomisation. Those outcomes cannot be converted into a trader's profit.

The populariser of the checklist outside medicine is Atul Gawande, one of the co-authors of that study and the author of a book about checklists[14]. The book is practical literature and is treated as such here: it supplies concepts and examples; there is no evidence in it.

The US Airways 1549 case – usually told back to front

On 15 January 2009 an Airbus A320 lost both engines after striking a flock of birds and ditched in the Hudson River. From the strike to the ditching, 208 seconds passed. There were 155 people on board and all of them survived. In the popular version of the story the captain "did not improvise, he ran the checklist from memory". The documented sequence looks different[9].

The crew reached for the paper emergency procedure list for loss of both engines. The list was written for a failure at cruise altitude and could not be completed at the altitude they were at. The auxiliary power unit was started earlier than the order on the list provided for, and it was precisely that departure that the report treated as material to how events unfolded. The conclusion for a trader is the opposite of the popular one and far more interesting: the procedure did not replace situational judgement – it took enough routine steps off the crew that there was room left for one non-standard decision. A good procedure does not remove thinking. It determines what to spend it on.

Design principles, without a ready-made list

Ready-made point sets, forms and the rhythm of the week are the content of the trader checklists appendix; what stays here are design rules you can apply to any list.

  • The list has to be short enough to be run at a fixed moment, rather than "when there is time".
  • The moment of execution must be defined by an event in the process: before exactly what, after exactly what.
  • The list is written down, not walked through in your head. Without a record there is no audit, and without an audit there is no way to tell whether it works.
  • Every point must have stopping power. A point that cannot block anything is decoration and takes up attention needed elsewhere.
  • Changes to the list happen after an event and after a review, never during a session.

The same list, written out in the four fields that apply across the whole section, looks like this:

The pre-session list as a procedure.
Trigger condition: opening the platform on a day you intend to trade.
Action: going through the points of the list in a fixed order, recording an answer at each one. Every point has stopping power.
Completion criterion: the last point has a recorded answer. The list is then closed and you do not come back to it during the session.
Audit trail: the completed form with the time of closing, plus the number of orders the list stopped that day.
Failure criterion (a requirement separate from the form): an order was sent despite an unmet point; no blocks calls for checking whether any were warranted.
A gate is a condition someone else can settle from your record 1. Condition on the list saved before the session? no → stop 2. Instrument on the list saved before the session? no → stop 3. Position size matches the record? no → recalculate 4. Protective order ready with the entry? no → set it 5. Rationale written before sending? no → write it order sent None of the conditions contains a threshold or a parameter — those come from the risk plan, not from this schematic.
The schematic shows the form of gating, not a ready-made list. Every condition is settleable from the record by a third party and every one has the power to stop the order.
A checklist as an order gate. Conditions and inferential limits are part of the diagram.
Format example, not a complete plan. Zero blocks may mean full compliance; test an example containing an unmet point.

The dead checklist

One possible checklist failure is completing it without distinguishing a compliant order from a noncompliant one. No blocked entries does not establish this by itself: every condition may have been met. Test the list against a prepared example with one unmet point and check whether it actually stops the decision. In the audit, look for orders sent despite a recorded failure, omitted fields, and lists completed only after entry.

A functioning list stops an order when a condition is unmet and leaves a checkable record. Irritation may occur, but it is not a measure of effectiveness. A list that never causes anger may work correctly; an irritating list may be poorly designed. Assess the condition and execution, rather than the emotional response.

9. When a procedure harms, or simply does not apply

This article's thesis has counterexamples and it is better to give them yourself than to pretend they do not exist.

First counterexample: a procedure executed correctly ends in a loss

An illustrative sequence, with no numerical values, because none are needed here. Before entry everything the procedure requires is recorded: condition from the list, instrument from the list, size matching the record, protective order submitted together with the entry, rationale entered before sending. Two endings are possible – price reaches the target level or it reaches the protective order.

The assessment of the process is identical in both cases, because it was settled before the outcome was known. Only the outcome changes, and the outcome of a single trade is one draw from a distribution. The reverse case is more interesting: an entry off the list, size off the record, no protective order – and a profit. The assessment of the process is settled there too and reads the same as after a loss: a violation. The profit is a cost here, not a reward, because it reinforces behaviour whose distribution on repetition looks nothing like that one payout. The full argument for separating decision quality from outcome quality is in article 10.4.

Second counterexample: a procedure imported from someone else's process

A rule closing every position before a listed data release makes sense in a process whose precondition is a quiet market – and is harmful in a process whose precondition is precisely the reaction to a release. The same rule, two opposite effects. A procedure is always a procedure of something. Copied without the process it belonged to, it does not become neutral; it starts constraining at random – and that is a real way in which good rules ruin good results.

Third counterexample: a barrier placed too late in the sequence

A safeguard standing at a point the sequence does not reach anyway is not failing – it simply does not apply. You will recognise this by three symptoms described more fully in article 10.5: the condition triggers more and more often while the number of violations does not fall; violations migrate to another instrument or another time of day; the barrier gets lifted mid-session. Each of those symptoms means the same thing – the safeguard has to be moved earlier, not accompanied by a repeated resolution.

Three boundaries of procedure effectiveness. Conditions and inferential limits are part of the diagram.
Conceptual diagram. It does not establish motive, cause or financial outcome.
A barrier does not protect against an event that jumps the threshold. A kill switch based on the current account state will work when it has time to: with a price gap, suspended quotes or a violent widening of the spread, the threshold is crossed and the crossing is already a fact before the mechanism closes anything. This is a shield for a bad day; it does not protect against an extreme event – and the difference is worth knowing in advance. The scope and construction of drawdown thresholds belong to section 9.
A limit on the number of positions is not a limit on exposure. Two positions on two strongly correlated instruments are one thesis with doubled risk, and the rule "at most two positions" cannot see that. A rule that counts units is easy to check and that is its only merit. Net exposure and how to measure it are described in section 9.

Worth noting separately is the case in which nobody held a server-side barrier at all. The collapse of Barings Bank in 1995 is documented by the report of the Board of Banking Supervision: losses came to around GBP 827 million – the widely repeated figure of 1.4 billion refers to US dollars[12]. The mechanics of the sequence and the conclusions for a retail account are discussed in article 10.5; here one thing counts: what failed there was not somebody's discipline, but the absence of a safeguard whose removal would have required somebody else's involvement.

10. Implementation, durability and the 21-day myth

"Twenty-one days is enough to build a habit" is a sentence that comes from no study at all. The number came from a popular book by a plastic surgeon in the 1960s, from observations of patients getting used to a changed appearance, and it was carried over to arbitrary behaviours by forty years of repetition.

A measurement does exist, and it looks different. Lally and colleagues followed simple daily behaviours in 96 volunteers for 12 weeks, or 84 days[10]. In well-fitting models, time to reach 95% of each individual’s automaticity asymptote had a median of 66 days and a range of 18–254 days. Longer times were model extrapolations beyond observation, rather than 254 days of participant follow-up. Measurement was self-reported, and the behaviours were simpler than a trading procedure. The finding challenges a universal 21-day deadline but sets no deadline for a trader.

How long automating a simple habit takes the popular "21 days" median 66 days 18 days 254 days 0 50 100 150 200 250 days from the start Lally et al. (2010): simple everyday behaviours, self-reported measure, modelled estimate of the spread.
Axis scale: 1.75 pixels per day, axis starting at zero. The band covers the 18–254 day spread, the vertical line marks the median of 66 days.

The practical conclusion is different from the one usually drawn. The point is not to prepare for 66 days instead of 21. The point is not to set yourself a deadline by which the procedure "is supposed to become automatic", because the absence of automaticity after a month is not information about you – it falls inside the spread. And something more important: a trading procedure does not have to become automatic in order to work. It is enough that skipping it is costly or visible.

In summary: how you will know a procedure has died

  • It did not stop an order despite an unmet condition.
  • It is run after the order instead of before it.
  • You settle its trigger condition by feel instead of from the record.
  • The rule is changed during the session instead of after a review.
  • The record is created, but nobody goes back to it.

Each of these points is checkable in data you need for other purposes anyway. None of them requires introspection.

11. What this article does not settle

A few things have deliberately not appeared here, and it is better to say plainly where to look for them than to leave the impression that the subject has been closed.

  • Numerical thresholds – daily and weekly limits, position size, drawdown levels: section 9. This article describes the form of a rule; the parameter comes from there.
  • Assessing whether a rule set makes sense – expected value, historical testing, judging an edge: section 8. A procedure protects an edge but does not create one.
  • Recording and process metrics – journal construction, violation codes, review cadence: article 10.8. Without it, no failure criterion from this text is checkable.
  • The interruption and return protocol – what exactly happens once a block is triggered and on what terms you come back: article 10.9.
  • Ready-made documents – the template a trading plan is built from, in the trading plan and playbook article, checklists and the rhythm of the week in the trader checklists appendix, the contingency plan in the emergency plan appendix.
  • Tool configuration – terminal settings and what is possible on the broker's side change quickly and belong to the platforms section.

One sentence remains, worth checking against your own record before you move on: a rule whose breach leaves no trace is not a rule – it is a memory of an intention. That is exactly why the recording side deserves its own reading: the trading journal (10.8) is where every failure criterion from this text becomes countable.

FAQ – designing trading procedures

Will writing a record before every order make me miss entries?
Sometimes; the delay cost must be checked for the particular condition and tool. If a signal disappears while a few fields are completed, either a short entry window or a slow form may be involved. No direct study establishing that this pause reduces spur-of-the-moment entries in retail FX is cited here. Across several dozen sessions, compare orders lacking a record with conditions that disappeared during completion. This describes cost and execution; it does not by itself establish the pause’s causal effect.
How do I tell whether my safeguard is hard rather than just a declaration?
Before the session, check who can remove the block, their permissions, how long it takes and whether bypasses are logged. Test limit removal, setting changes, access to funds and any required second-person approval. Rapid removal indicates easy reversibility; comparing the time with order placement is not a scientific threshold. Hardness comes from the actual restriction of options, rather than server location alone. A soft reminder may help, but should not be credited with a technical block it does not provide.
How long does it take before a procedure starts working?
There is no universal deadline. In the Lally study, the modelled time to 95% of the asymptote of self-reported automaticity had a median of about 66 days and a range of 18–254 days; 254 days was not a direct observation period. Simpler everyday behaviours were studied, rather than trading procedures. The finding justifies neither a "21-day" rule nor replacing it with 66. A procedure can be executed consciously before becoming habitual. Visible omission and bypass costs help assess it, but do not themselves guarantee operation.
Will rigid rules take away my ability to react to unusual situations?
They may constrain a response; that cost belongs in the design. A broad "unless the situation is exceptional" leaves considerable room for bypassing. A better starting point is to prepare conditions and actions: no quotes for a defined period, instrument suspension, a specified spread widening or a listed release. A closed list alone does not guarantee safety; it must cover relevant events and specify the response to an unforeseen failure. Parameters and technical execution need separate checking.
Can I base a rule on how I feel before the session?
A scale can be a useful declaration if the question, level descriptions and entry timing stay fixed. A previously recorded rule may link a particular answer to an action, but review whether ratings are being shifted to avoid the restriction. A checkable event—missing fields, a listed violation or a specified situation—therefore often gives a stronger audit trail. Self-report is neither a diagnosis nor proof of a behavioural cause.
Will procedures improve my results?
Not in the sense in which that is usually meant. A procedure does not raise hit rate, does not improve the signal and does not create an edge – it protects a process that already has a positive justification from erosion in execution. If that justification is absent, an orderly way of losing money remains a way of losing money. Assessing whether a rule set makes sense is the subject of section 8. There is also the opposite trap worth guarding against: a procedure is not judged by the outcome of a single trade or even by a good month, because the assessment of the process was settled before the outcome was known.
How often may I change my own rules?
Not during a session and not after a single event, because a rule changed in the middle of an episode is being changed by that episode. Beyond that there is no universal cadence – any number I gave here would be invented. A qualitative criterion makes more sense: a change follows a review of a run in which the event in question occurred repeatedly, not a single instance. The exception is when a review reveals the absence of a rule for an event that has already happened and was costly – then you add it immediately, but outside the session.
How do I know that a particular rule has stopped working?
Compare its purpose, trigger opportunities and actual execution. No stopped orders may mean no qualifying conditions, rather than failure. Review flags include late action, ambiguous condition assessment, an unauthorised bypass and records that are never reviewed. Some are visible in system history; others require a journal. Establish what failed: recognition, enforcement, coverage or return. The failure criterion concerns the intended purpose, rather than treating every absence of blocking as a failure.

Sources and references

  1. Loewenstein G., "Out of Control: Visceral Influences on Behavior", Organizational Behavior and Human Decision Processes, 65(3), 1996, pp. 272–292. The cold–hot empathy gap: a systematic error in predicting one's own behaviour in a different state of arousal. Laboratory studies, student samples, non-financial stimuli – what is carried over to trading is the mechanism, not the effect size.
  2. Gollwitzer P.M., "Implementation Intentions: Strong Effects of Simple Plans", American Psychologist, 54(7), 1999, pp. 493–503. Introduction of the if–then plan concept and its format.
  3. Gollwitzer P.M., Sheeran P., "Implementation Intentions and Goal Achievement: A Meta-Analysis of Effects and Processes", Advances in Experimental Social Psychology, 38, 2006, pp. 69–119. Meta-analysis of 94 independent studies, more than 8,000 participants, effect size d ≈ 0.65 for acting on adopted goals. Goals mainly health, educational and consumer; no studies on traders.
  4. Baumeister R.F., Bratslavsky E., Muraven M., Tice D.M., "Ego Depletion: Is the Active Self a Limited Resource?", Journal of Personality and Social Psychology, 74(5), 1998, pp. 1252–1265. The original formulation of the model of self-control as a limited resource; laboratory experiments on students, symbolic stakes.
  5. Hagger M.S. et al., "A Multilab Preregistered Replication of the Ego-Depletion Effect", Perspectives on Psychological Science, 11(4), 2016, pp. 546–573. A preregistered replication report: 23 laboratories, N = 2,141, a common protocol. Effect d ≈ 0.04, confidence interval covering zero – indistinguishable from zero. This is not a meta-analysis and the result is not "mixed".
  6. Danziger S., Levav J., Avnaim-Pesso L., "Extraneous Factors in Judicial Decisions", Proceedings of the National Academy of Sciences, 108(17), 2011, pp. 6889–6892. Observational data, about 1,100 decisions of Israeli parole boards, no random assignment of cases.
  7. Weinshall-Margel K., Shapard J., "Overlooked Factors in the Analysis of Parole Decisions", Proceedings of the National Academy of Sciences, 108(42), 2011, E833. A methodological critique of the above: the order of cases was not random, which leaves an alternative explanation of the observed pattern open.
  8. Haynes A.B. et al., "A Surgical Safety Checklist to Reduce Morbidity and Mortality in a Global Population", New England Journal of Medicine, 360(5), 2009, pp. 491–499. Eight hospitals in eight cities, about 3,700 patients before implementation and about 4,000 after. Mortality 1.5% → 0.8% (a fall of 0.7 percentage points), complications 11.0% → 7.0% (a fall of 4.0 percentage points). A before-and-after comparison, without randomisation and without a control group.
  9. National Transportation Safety Board, Loss of Thrust in Both Engines After Encountering a Flock of Birds and Subsequent Ditching on the Hudson River, US Airways Flight 1549, Airbus A320-214, January 15, 2009, aircraft accident report NTSB/AAR-10/03, Washington 2010. The chronology of the event, the use of the emergency procedure list and the earlier-than-scheduled start of the auxiliary power unit.
  10. Lally P., van Jaarsveld C.H.M., Potts H.W.W., Wardle J., "How Are Habits Formed: Modelling Habit Formation in the Real World", European Journal of Social Psychology, 40(6), 2010, pp. 998–1009. Simple everyday behaviours, self-reported measurement over 12 weeks; median time to an automaticity plateau about 66 days, modelled spread about 18–254 days.
  11. Ariely D., Wertenbroch K., "Procrastination, Deadlines, and Performance: Self-Control by Precommitment", Psychological Science, 13(3), 2002, pp. 219–224. Students, coursework; imposed deadlines produced better results than self-imposed ones, and those in turn better than a single final deadline.
  12. Board of Banking Supervision, Report of the Board of Banking Supervision Inquiry into the Circumstances of the Collapse of Barings, HMSO, London, 18 July 1995. The primary source for the size of the losses (about GBP 827 million) and for the description of the absence of independent control on the institution's side.
  13. Odean T., "Are Investors Reluctant to Realize Their Losses?", Journal of Finance, 53(5), 1998, pp. 1775–1798. About 10,000 brokerage accounts, US equity investors, 1987–1993 – empirical documentation of the disposition effect, referenced in this course in article 10.6. The concept was introduced by Shefrin and Statman (1985); carrying it over to the leveraged retail currency market is an extrapolation.
  14. Gawande A., The Checklist Manifesto: How to Get Things Right, Metropolitan Books, 2009. Practical literature: a source of concepts and examples concerning checklists, not of empirical evidence.
  15. Vohs K.D. et al., “A Multisite Preregistered Paradigmatic Test of the Ego-Depletion Effect”, Psychological Science, 32(10), 2021, pp. 1566–1581. 36 laboratories, N = 3,531, d = 0.06; the main effect was non-significant in this protocol. Full text.
  16. Dang J. et al., “Revisiting Ego Depletion: Evidence from Multi-Lab Collaborations”, Journal of Pacific Rim Psychology, 2025. DOI: 10.1177/18344909251386084.

Your learning progress

Finished this lesson? Mark it yourself. Opening or scrolling a page never marks it complete.

Progress requires JavaScript and browser storage.

Saved only in this browser. No account or cross-device sync. Clearing site data can erase your progress.

Jarosław Wasiński

Editor-in-chief, ForexMechanics.com • Macro & FX analyst

Independent analyst and practitioner with over 20 years of experience in financial markets. Actively involved in Forex since 2007, with a focus on fundamental analysis, OTC market structure, and disciplined capital risk management.

  • Creator and editor-in-chief of MyBank.pl, providing reliable financial education since 2004.
  • Author of hundreds of in-depth market commentaries, structural analyses, and educational materials for traders and investors.
  • Advocate for market transparency, promoting education grounded in hard data and institutional reports (BIS, ESMA, CFTC).

All content is educational and informational only – it does not constitute investment advice or a recommendation. Remember: leveraged trading (Forex/CFD) carries a high risk of rapid capital loss.