Get Out of Your Own Way
A practicum in personal agency: the observation, the decision and the repeated action that move a stuck situation, built on what the evidence actually supports.
- Levels
- 5
- Lessons
- 27
- Knowledge checks
- 55
- Gates
- 38
- Working tools
- 8
- Simulations
- 1
Most people who are stuck already know what to do. They know they should apply, prepare, speak, finish, ask, decline, or confront. Knowing has not produced the action, and no further explanation of why the action matters is going to produce it either.
That gap is measurable. In the largest meta-analysis of procrastination, the correlation between how much people intend to work and how much they delay is 0.03 across eight studies and 1,017 people (Steel, 2007). Steel's own summary is that procrastinators intend to work as hard as anyone else. Intelligence correlates with delay at 0.03 as well. Whatever is holding the position, it is not information and it is not intent.
This is not a motivation course. What it builds instead is a small set of observations you can make about yourself, a decision procedure you can run when you do not want to, and a record of completed actions that eventually becomes the only evidence about yourself worth trusting.
The course runs on one subject: a real situation in your own life or work that has been unresolved for at least three months. You choose it before Level 0 ends and keep it for the whole course.
What this kind of course has been shown to do
There is one field trial that tests something close to what this course teaches, and one that fails to reproduce it. You should know both before Level 0.
In Lomé, Togo, 1,500 microenterprise owners were randomly assigned to three groups of 500: conventional business training covering accounting, marketing and finance; personal initiative training covering self-starting behaviour, goal setting, planning and overcoming obstacles; or nothing. Both training arms received the same 36 classroom hours and the same four monthly follow-up visits. Over 29 months, the personal initiative group raised monthly profits by 30 per cent. The conventional business training group raised them by 11 per cent, which was not statistically distinguishable from zero (Campos et al., 2017).
Two years later, several of the same researchers ran the same family of training with 2,001 women entrepreneurs in Addis Ababa. At 18 months there was no effect on profits, no effect on business practice, and no effect on the psychological measures the training was designed to move (Alibhai et al., 2019). The authors' explanation was the trainers: only 41 per cent had ever run a business themselves.
Set both against the base rate. Across 37 evaluations of entrepreneurship and initiative programmes in 25 countries, producing 1,116 estimates, 68 per cent of estimates were statistically insignificant (Cho & Honorati, 2014). And when behavioural interventions move from research trials to delivery at scale, effects fall by roughly a factor of six, which publication bias and low statistical power fully account for (DellaVigna & Linos, 2022).
So: this material has produced a large, durable, replicated-in-one-city effect on people's actual income, and a clean null in another city eighteen months later, inside a literature where two thirds of estimates are null.
Trained owners borrowed more money than untrained ones. They were not more likely to be granted a loan. Initiative moved what was under their control, which was demand for capital. It did not move what was not, which was the supply of it.
That sentence is the scope of this course, stated by its own best evidence.
The five plateaus
You will move through five capabilities, each gated by behaviour someone else could watch you fail.
First you learn to describe your situation accurately, separating what happened from what you concluded and what you are now avoiding. Then you decide what you are willing to pay for the result you say you want. Then you act while the discomfort is present and before the information is complete. Then you interrupt the cycle that turns one missed day into an abandoned commitment, and start accumulating completed actions. Finally you build a system that runs in a bad week, and prove it over thirty days.
Cadence
Plan for three to four hours each week: about an hour of lessons and checks, two hours on the exercise against your own subject, and twenty minutes on the weekly review. Levels 0 to 3 take two weeks each. Level 4 takes four, because the final project is thirty days of dated evidence.
Where the evidence comes from, and where it does not
Almost every psychological finding in this course was established on a Western sample. In an audit of top journals, 96 per cent of samples came from countries holding 12 per cent of the world's population, and 68 per cent came from the United States alone (Henrich, Heine & Norenzayan, 2010). Seven years later, 94 per cent of studies in one flagship journal still sampled Western countries, and over 91 per cent of studies reported nothing at all about their participants' socioeconomic position (Rad, Martingano & Ginges, 2018).
Most of the procrastination literature specifically was built on undergraduates facing clustered institutional deadlines, with no salary, no manager and no colleague depending on them. Workplace procrastination shares only about a fifth of its variance with the trait that literature measures, and its strongest correlate is boredom driven by insufficient job demands (Metin, Taris & Peeters, 2016).
None of this makes the findings false where you are. It makes them untested where you are, and this course says so at each point rather than once at the beginning.
What this course does not claim
Some situations are structural, and in Nigeria the structure is documented. On the official labour force survey, 93 per cent of employment is informal, 85.6 per cent of workers are self-employed, and only 14.4 per cent hold wage employment (National Bureau of Statistics, Q2 2024). Unemployment among people with post-secondary education is 4.8 per cent against 2.3 per cent among people with no formal education, which is the opposite of the direction most people expect. The World Bank puts primary wage employment at 13.6 per cent and poverty at 63 per cent of the population in 2025, with 3.5 million people entering the labour force each year.
The salaried professional job a reader may be organising their life around is, statistically, a category that mostly does not exist. Between 2014 and 2018 the share of Nigerian youth planning to leave the country permanently rose from 36 per cent to 52 per cent, among the highest in sub-Saharan Africa.
The working position of this course is narrower than either "everything is in your control" or "nothing is": in many stuck situations there is a portion under your control, it is usually larger than your current behaviour reflects and much smaller than you were sold, and how large it is varies enormously by context. The identical training produced 30 per cent in Lomé and zero in Addis Ababa, and nobody knew which it would be until the endline.
Across 64 studies and over 12,000 participants, judging that someone's situation was under their control reliably reduces sympathy and help, and increases hostility toward them (Rudolph et al., 2004). That applies to you judging yourself after a failure that was in fact structural.
This course raises your estimate of how much is controllable. That estimate has an error rate, and its errors are asymmetrically expensive to you. When a specified action, taken for a specified period, produces no specified result, believe the record rather than the framework.
A level is passed when you have done something and can show what happened, not when you agree with the lesson. Reading this page changes nothing. It is not supposed to.
The levels
- 0See it accuratelyReplace the explanation you give for being stuck with a description someone else could check.Free
- 1Decide what you will payConvert one want into a costed commitment, or move it to a shelf with a review date.Locked
- 2Act while it is uncomfortableComplete the action with the fear present and before the information is complete.Locked
- 3Recover and accumulateInterrupt the abandonment sequence at a named point, and build a record of completed actions.Locked
- 4Run the systemOperate a personal system that survives a bad week, and produce thirty days of dated evidence on one real problem.Locked
Stuck subject
One real situation, unresolved for three months or more, carried through every exercise in the course.
Before you finish Level 0 you choose the subject. It stays fixed for twelve weeks.
There is a specific reason to hold to one. Forming detailed plans improves success on a single goal and stops helping, or actively reduces commitment, when it is applied to several at once. In one set of experiments, planning helped people pursuing one goal and produced no benefit for people pursuing six, because planning surfaced how much execution the whole set required and commitment fell (Dalton & Spiller, 2012). Goal conflict is also robustly associated with distress across 54 samples at a median correlation of about 0.34, and in the one behavioural study people took less action on conflicted goals while thinking about them more (Emmons & King, 1988).
Changing subject halfway is the most common way this course fails, because the second subject is always chosen at the moment the first one starts to cost something.
What qualifies
- It is real. It involves named people, actual dates, a document, an application, a conversation, a sum of money, or a piece of work with a deadline.
- It has not moved in at least three months.
- You have described it to someone else at least once, using an explanation that felt true when you said it.
- There is at least one action available to you this week that you have not taken.
- The outcome matters enough that you would be annoyed to read this page in a year and find it unchanged.
What does not qualify
A goal you have never attempted does not qualify, because there is no avoidance pattern to examine yet. A situation entirely controlled by another person does not qualify, though the part where you have not asked them anything does. A subject you would be unwilling to write about honestly does not qualify.
Examples of qualifying subjects
The professional certification you have discussed for two years and prepared for on three weekends. The internal role you did not apply for. The manager you have not asked for feedback. The service you have designed, named and priced but never offered to anyone. The report at eighty per cent that has been at eighty per cent since March. The conversation with a colleague you have rehearsed eleven times and never had.
If two subjects come to mind and one of them makes you slightly reluctant to write it down, that is the one with the pattern in it.
Recording it
Write the subject in one sentence that names the person or organisation involved, the outcome you want, and the date it stopped moving. If you cannot write it without qualifications, the subject is still an area rather than a situation, and it needs narrowing.
Rules of practice
Six rules. Four are supported by evidence, two are design choices, and the difference is marked.
1. Everything is written and dated
Evidence: moderate. Across 138 randomised trials and 19,951 people, prompting someone to monitor their progress raised actual goal attainment by d = 0.40, and the effect was larger when progress was physically recorded than when it was tracked without recording: 0.43 against 0.29 (Harkin et al., 2016). Correcting for publication bias pulls the overall figure down to about 0.19, so the honest range is 0.19 to 0.40.
Recording is roughly a third more effective than not recording. It is not the difference between something and nothing. What it does reliably is stop you remembering the weeks that went well.
2. Nothing here requires you to feel ready
Evidence: strong, by analogy. Behavioural activation treats depression by scheduling and completing activities without first repairing mood or belief. Against controls it produces a standardised mean difference of 0.74 across 26 trials (Ekers et al., 2014), and in a 440-patient trial it was non-inferior to full cognitive behavioural therapy at twelve months while costing about 21 per cent less (Richards et al., 2016). Acting first is not a slogan borrowed from motivational writing; it is a treatment that matches the alternative that works on beliefs first.
3. The smallest real action beats the planned large one
Evidence: mixed, and weaker than it is usually stated. See lesson 1.5. What survives is narrower: an action that leaves your possession creates information that an action inside your own head cannot.
4. Missing is information, not a verdict
Evidence: strong. This is the best-supported claim in the course and Level 3 is built on it.
5. Discomfort travels with you
Evidence: strong. In exposure research, how much fear falls during an exercise does not predict whether the exercise worked. Interventions repeatedly move behaviour and physiology while leaving self-reported fear unchanged (Kircanski et al., 2012; Niles et al., 2015). Feeling better during the action is an unreliable measure of whether the action did anything.
6. You claim your own gates
Design choice, not evidence. No one marks these for you.
One lesson block read, the knowledge checks answered wrong at least once, the exercise run against the real subject, three dated lines in the record, and one thing you avoided named honestly.
All lessons read, all checks correct on the first attempt, the exercise done in your head, the record blank, and a private sense that you understood it.
Assessment
What counts, what does not, and what you should be able to show at the end.
What is assessed
| Component | Weight | The evidence |
|---|---|---|
| Exercise records | 20% | Dated entries against the real subject, one per exercise |
| Weekly actions | 20% | Completed actions with outcomes recorded, including the failures |
| Recovery behaviour | 15% | What happened in the twenty four hours after each missed action |
| Accountability reviews | 15% | Dated exchanges with a named partner, with what they challenged |
| The thirty day project | 30% | The daily record, four weekly reviews, and the final report |
What is not assessed
Confidence is not assessed, and lesson 3.4 explains why the evidence says it should not be. Fluency in discussing your patterns is not assessed. Insight is not assessed.
A measurement warning that applies to your own self-assessment
Ten self-report procrastination instruments were tested against actual behaviour in one study of 235 people. The best predicted days-to-completion at r = 0.19, several predicted it not at all, and none predicted the pacing style most associated with procrastination (Vangsness et al., 2022). Three separate experience-sampling studies found that baseline trait self-reports did not explain individual delay behaviour, while momentary judgements about a specific task did.
Your account of yourself as a procrastinator correlates with your actual delay at roughly 0.18. This is why every exercise in this course asks for dated records of specific acts rather than for ratings of your tendencies.
The scorecard
Ten statements, each scored one to five, at the start of the course and again at the end. Run it before Level 0 and record the date. The absolute number means little and the change on individual lines means more, particularly acting without motivation, recovering after mistakes, and deciding without complete certainty.
What you should hold at the end
A named subject that has moved, with the movement dated. A record of at least twenty completed actions, including the ones that produced nothing. A written pattern map of your own avoidance sequence. A recovery rule you have used at least twice. Four weekly reviews. One final report in which every claim about what you learned is backed by a dated entry.
Sources
Every figure in this course, with the study behind it and what it can and cannot support.
Full citations with links are in SOURCES.md in the course directory. This page is the map: what each level rests on, and how strong it is.
How to read the grades
Strong means large samples, randomised or prospective designs, and either replication or a bias-corrected estimate. Moderate means well conducted but single-source, or meta-analytic with known publication bias. Weak means small samples, one laboratory, or a claim that circulates without a controlled test behind it.
Where a figure is weak, the course says so in the lesson rather than here.
Level 0, on observation
| Finding | Figure | Grade |
|---|---|---|
| Intention is unrelated to delay | r = 0.03, 8 studies, N = 1,017 (Steel, 2007) | Strong |
| Perfectionism is unrelated to delay | r = 0.04, 24 studies, N = 3,884 | Strong |
| Interpretation of a break drives disengagement | 53.2% intact, 42.0% external cause, 28.9% self-attributed, N = 418 randomised (Silverman & Barasch, 2022) | Strong |
| Self-criticism harms goal progress; standards help | d+ = −0.54 and +0.27, five prospective samples, objective outcomes (Powers et al., 2011) | Strong |
| Self-report of delay barely tracks actual delay | best instrument r = 0.19, N = 235 (Vangsness et al., 2022) | Moderate |
| Delay responds to environmental reliability | 182 s vs 722 s, N = 28 (Kidd, Palmeri & Aslin, 2013) | Moderate, small sample |
| Marshmallow prediction shrinks under controls | β 0.236 → 0.081 → 0.050, N = 918 (Watts, Duncan & Quan, 2018) | Strong |
| Opportunity cost neglect | purchase rate 75% → 55% from a semantic restatement (Frederick et al., 2009) | Moderate |
Level 1, on cost and commitment
| Finding | Figure | Grade |
|---|---|---|
| Positive fantasy predicts less attainment; expectation predicts more | r = −0.16 to −0.43 and +0.21 to +0.55, four domains (Oettingen & Mayer, 2002) | Moderate, correlational |
| Difficult goals beat easy goals | d = 0.45 vs 0.25, 384 cases, N = 16,523 (Epton, Currie & Armitage, 2017) | Strong |
| A demanding goal with permitted, costly skips beats both | 52.5% vs 25.9% vs 21.1% (Sharif & Shu, 2017) | Moderate |
| Rewarding return after a miss was the best of 53 interventions | +27%, N = 61,293 (Milkman et al., 2021) | Strong |
| Implementation intentions, bias-corrected | d = 0.15 from 642 tests; self-generated cues 0.16 (Sheeran, Listrom & Gollwitzer, 2024) | Strong |
| Event anchors vs time anchors, head to head | no difference, preregistered RCT, N = 192 (Keller et al., 2021) | Strong |
| Planning prompts for repeated gym attendance | null, CI excludes any gain above 2%, N = 877 (Carrera et al., 2018) | Strong |
| Enforced routine produced weaker habits than flexibility | N = 2,508 (Beshears et al., 2021) | Strong |
| Shelving a goal matches abandoning it, with less regret | N = 214 randomised (Mayer & Freund, 2022) | Moderate |
| Planning across six goals lowers commitment | N = 67, 216, 107 (Dalton & Spiller, 2012) | Weak to moderate |
| Time to automaticity | median 66 days, range 18 to 254 (Lally et al., 2010); replications 59 to 66 | Moderate |
| Commitment devices: take-up and failure | 11% to 42% take-up; 55% of takers default (Giné et al., 2010; John, 2020) | Strong |
Level 2, on acting under discomfort
| Finding | Figure | Grade |
|---|---|---|
| Behavioural activation against controls | SMD = 0.74, 26 RCTs (Ekers et al., 2014); non-inferior to CBT, N = 440 (Richards et al., 2016) | Strong |
| Acceptance-based vs cognitive approaches | g = −0.01, p = 0.86, 34 RCTs | Strong |
| Fear reduction during exposure does not predict outcome | replicated null (Craske et al., 2012) | Strong |
| Interventions move physiology while self-reported fear does not change | Kircanski et al., 2012; Niles et al., 2015 | Moderate |
| Naming dampens positive feeling too, and works for non-emotional labels | Lieberman et al., 2011; Constantinou et al., 2014 | Moderate |
| People underestimate compliance with requests | by about 48%, 12 studies, 14,000+ targets (Bohns, 2016) | Strong |
| Asking for advice raises perceived competence, unless the advisor is a non-expert | d = 0.35 to 0.74; non-expert d = 0.41 below asking nobody (Brooks et al., 2015) | Moderate |
| Reversible outcomes are less satisfying, and people prefer them anyway | 66.3% choose changeable (Gilbert & Ebert, 2002) | Moderate |
| Information avoidance is common | 18% never collect HIV results; 40.4% avoid a doctor (Golman et al., 2017) | Strong as review |
| Choice overload | D = 0.02, CI spanning zero, 63 conditions (Scheibehenne et al., 2010) | Strong |
| Cost of delay in a job search | callbacks 45% lower at 8 months, ~12,000 applications (Kroft et al., 2013) | Strong |
| Planning fallacy | predicted 33.9 days, actual 55.5; ~30% finish on time (Buehler et al., 1994) | Moderate |
| Durability bias operates over months to years, not days | tenure denial, significant at 1 to 5 years, closed by 6 to 10 (Gilbert et al., 1998) | Moderate, disputed magnitude |
Level 3, on recovery and evidence
| Finding | Figure | Grade |
|---|---|---|
| Repair option after a broken streak | 85.2% vs 68.7%, OR 2.63 (Silverman & Barasch, 2022) | Strong |
| One missed opportunity costs almost nothing | −0.29 SRHI points, not significant, no lasting cost (Lally et al., 2010) | Moderate |
| Self-efficacy is mostly a product of past performance | forward ρ = 0.01 with controls, backward ρ = 0.32, N = 34,870 (Sitzmann & Yeo, 2013) | Strong |
| The same asymmetry in academic data | 0.205 backward vs 0.071 forward (Talsma et al., 2018) | Strong |
| Bolstering self-esteem in struggling students backfired | 57% → 38% on the final exam, N = 86 (Forsyth et al., 2007) | Moderate |
| Shame predicts withdrawal; guilt predicts repair | Tangney et al., 2007; shame-proneness → procrastination, guilt r = −0.08 ns (Martinčeková & Enright, 2018) | Moderate to strong |
| One intense session non-inferior to 4 to 20 graded sessions | SMD = −0.123, N = 268 (Wright et al., 2023) | Strong |
| Graded vs variable exposure intensity | no outcome difference; 5 withdrawals vs 0, N = 40 (Jacoby et al., 2019) | Weak to moderate |
| Coping plans add nothing to action plans | 0.41 vs 0.30, 35 RCTs, N = 5,439 (Liang et al., 2022) | Moderate to strong |
| Procrastination interventions, RCTs only | g = 0.34, CI 0.11 to 0.56 (Rozental et al., 2018) | Moderate |
Level 4, on systems and accountability
| Finding | Figure | Grade |
|---|---|---|
| Monitoring progress raises attainment | d+ = 0.40, CI 0.32 to 0.48, k = 138, N = 19,951; 0.19 bias-corrected (Harkin et al., 2016) | Strong |
| Recording beats not recording | 0.43 vs 0.29 | Moderate |
| Reporting to a person beats private monitoring | 0.47 vs 0.19; public 0.55 | Moderate to strong |
| Monitoring moves only what it measures | behaviour 0.79 / outcomes 0.14; outcomes 0.62 / behaviour 0.17 | Strong |
| Monitoring interval was never tested | not a moderator; duration null at β = 0.00 | Strong absence |
| Information-carrying vs reinforcing feedback | d = 0.99 vs 0.24, 435 studies, N > 61,000 (Wisniewski et al., 2020) | Strong |
| Over a third of feedback interventions harm performance | d = 0.41, 38 per cent of 607 effects negative, N = 12,652 (Kluger & DeNisi, 1996) | Strong |
| Accountability partners as an isolated component | one RCT, N = 69, no difference between arms | Weak; essentially untested |
| Over-specified plans halve the effect | 0.46 for when and where; 0.24 adding how and how long | Moderate |
| Effects that survive the intervention period | 8% of 53 conditions, N = 61,293 (Milkman et al., 2021) | Strong |
Scope and generalisability
| Finding | Figure |
|---|---|
| Personal initiative training, Togo | +30% monthly profits, p < 0.01, 29 months, N = 1,500 (Campos et al., 2017) |
| The same family of training, Ethiopia | no effect on profits, practice or psychological measures, N = 2,001 (Alibhai et al., 2019) |
| Entrepreneurship programme base rate | 68% of 1,116 estimates insignificant, 25 countries (Cho & Honorati, 2014) |
| Research trials vs delivery at scale | 8.7 pp vs 1.4 pp, 126 RCTs, 23 million people (DellaVigna & Linos, 2022) |
| Sample composition in psychology | 96% of samples from countries holding 12% of the population (Henrich et al., 2010); 94% still Western in 2014 (Rad et al., 2018) |
| Nigerian labour structure | 93.0% informal, 85.6% self-employed, 14.4% wage employment (NBS, Q2 2024); 13.6% primary wage work, 63% poverty in 2025 (World Bank) |
| Attributing control reduces help and increases hostility | 64 studies, 12,000+ participants (Rudolph et al., 2004) |
Claims removed from this course, and why
"95 per cent if you have an accountability appointment." Attributed to a training association. No author, no sample, no method, no journal, and absent from that organisation's own database and every scholarly index. Replaced by 0.47 against 0.19.
"42 per cent more likely if you write goals down." Does not appear in the study it is attributed to, which reported self-rated means of 6.44 against 4.28, and in which one group scored lower after adding action commitments.
The 1953 Yale written-goals study. Does not exist.
"21 days to form a habit." From a 1960 book by a cosmetic surgeon describing how long patients took to stop seeing their old face. He said minimum, and he was not describing habits.
"Inaction regret dominates over time." Failed to reproduce in the original author's own public replication with 2,600 people: long-term action 51 per cent against inaction 49 per cent, p = 0.35.
"The what the hell effect." The coining reference is a book chapter with an illustrative vignette. There is no study in it.
"Relief arrives within seconds when you avoid." Experience sampling finds guilt rather than pleasure at the moment of avoidance, and the lagged evidence runs from mood to delay rather than delay to relief.
What is open, and what is not
The free levels of this programme are readable with no account at all — the real levels, not samples. An account carries your progress, your gate claims and your saved work. The remaining levels, the tools, the gates and this programme’s worked exemplars are opened together when you enrol.
Create an account