Home › Patterns › Starting Over From Scratch
Why do I scrap good work and start again from nothing?
The short answer
Because the version in front of you carries the record of every compromise you made, and you are the only person who can see them. A blank page removes all of it at once, and the relief is immediate. Sometimes the judgment is correct and the work really was wrong. The test is whether anyone but you can tell the versions apart.
People recognize this in one specific sentence rather than in the general description, so start with these.
- "I deleted four thousand words that were fine and started again at zero."
- "I am on version eleven. Nobody else can tell version eleven from version two."
- "The only part I enjoy is the first three hours. After that I want to start something else."
- "I know exactly what is wrong with it. Nobody who has read it has ever mentioned it."
How the seminar works with what the restart is actually doing for you
Here is the moment this page is about. It is eleven at night, the document has about four thousand words in it, and you have been reading it from the top. Somewhere around the third page you hit the paragraph you already knew was weak, the one you wrote around a problem rather than through it, and the whole document goes gray. Every sentence after that reads as damaged. And then a very calm sentence arrives, with no drama attached to it at all: it would be faster to just do this again properly.
You do not usually press delete. You rename the file, or you make a new one, and the old one goes into a folder with a name like old or v1 or archive, which feels like caution and functions like a bin, because you will not open it again. Then you type a new first line. And for about twenty minutes it is the best you have felt about this project in two weeks.
There is a version of the same evening that involves no writing at all. The design was nearly right. Two elements were fighting each other and one color was wrong and the spacing on the second panel had been bothering you since Tuesday. Fixing those three things is roughly two hours of dull, precise work. Starting again is roughly two hours of interesting work, and the interesting version wins at nine in the evening. There is another version with a spreadsheet in it, where something in the totals is off by a small amount, and finding out where would mean reading your own formulas backward through eleven columns, and rebuilding the whole sheet from the source data feels cleaner and takes the rest of the afternoon.
And there is the version where none of this happens once. You are on version eleven. A colleague looked at version four and said it was ready. You know why it was not. You have been working on this for five months, and if somebody laid all eleven versions on a table and asked a person who had not seen them to sort them into an order, they could not do it.
Four boundaries help before going further, because several different behaviors get typed into a search bar in almost the same words and they lead to genuinely different places. If the work reaches the edge of finished and stops, intact, and sits in a folder, that stop near the end has its own page. If nothing ever begins because there is always more to read first, the preparation loop is the subject there. If the thing being restarted is a plan rather than an object, and the real start has moved to next Monday for the fourth time this year, the broken week and its reset are covered separately. If it happens to outgoing messages, four lines taking thirty-five minutes, that has a different mechanism and its own page. This page stays with one narrow thing, which is completed work being destroyed so that it can be made again.
What Does the Moment Before You Delete It Actually Look Like?
Undramatic, and that is the first surprise for most people.
There is no frustration in it. There is no throwing anything across a room. What there is instead is a quiet arrival of clarity, and clarity is a very persuasive feeling to have at eleven at night. The thing in front of you stops looking like a piece of work with three problems in it and starts looking like a piece of work that is wrong all the way down. The three problems have not multiplied. Your reading of them has changed.
Watch the specific sequence, because it repeats with unusual reliability. You reread from the beginning. You pass the section you already know is weak. Everything after it now reads through that weakness. By page four you are not evaluating the sentences on page four at all, you are still evaluating page two, and the verdict has spread. Then comes the sentence that does the actual damage, which is that starting again would be faster. That estimate arrives with total confidence and it is drawn entirely from the first twenty minutes of the last new file, which were indeed fast, and not at all from the four thousand words that took eleven days.
The physical action is worth noticing too, because it is almost never destruction in the obvious sense. You keep the old file. You name it something respectful. Keeping it feels like an act of prudence, and it produces the same outcome as deletion in about ninety percent of cases, because the new one starts from a blank page rather than from the old one open beside it. The old version has been retired without ever being consulted again, which is a quieter thing than deleting and a nearly identical thing in its effects.
One more feature belongs here, and people recognize it with a jolt. You are fast at beginnings. The first three hours of anything you make are noticeably better than most people's first three hours, and you know that, and it is true. That fluency comes from repetition. Somewhere between five and fifty times, you have practiced the opening of this kind of work far more than you have practiced its ending.
Why Does the Blank Page Feel Better Than the Nearly Finished One?
Because a blank page carries no history, and everything you have already built carries all of it.
Consider what is actually stored in a nearly finished object. Not just the words or the shapes, but the record of how each one got there. The paragraph on page two that you wrote around a problem instead of through it. The element you moved four times and settled on the fourth position without being convinced. The formula that gives the right answer through a route you would not defend out loud. None of that is visible in the object. All of it is visible to you, because you were there for every decision and the losing options stayed in your memory, attached.
That is why the object looks worse to you every time you read it. Each reading refreshes the record. A stranger reads the surface, once. You read the surface plus the archaeology, over and over, and the archaeology never improves.
A new file has none of it. That is the whole of its appeal, and the appeal has nothing to do with a forecast about quality. A blank page is flawless for the simple reason that it is empty, and the flawlessness holds for about twenty minutes, until the first compromise gets recorded and the process starts again.
The economics are worth stating plainly because they explain the durability of the habit. The relief of the new file arrives instantly, requires no cooperation from anybody, and costs nothing at the moment of purchase. The bill lands weeks later, on a person who is now on version seven and cannot remember what version three looked like. Two events that far apart do not connect in ordinary experience, which is how a capable person can run this for a decade and describe it, accurately, as caring about the quality of their work.
What Do Researchers Call This, and Where Do the Names Stop?
There is no single name, and the four or five bodies of work that touch it each describe the act with real precision while leaving your particular reason completely untouched.
The oldest useful vocabulary comes from research on revision, and specifically from the finding that the word means two different things to two kinds of people. Sommers interviewed student writers and experienced adult writers in 1980 and found the students described revision as changing words, cleaning up, and doing the thing over again. Their attention sat at the level of vocabulary. Experienced writers described something else entirely, which was finding the shape of the argument inside the draft they already had, and they worked in cycles rather than in passes. Perl reported a related pattern in 1979, watching unskilled writers interrupt themselves to correct surface features so early and so often that the meaning never had room to develop. Faigley and Witte built a taxonomy in 1981 that separates surface changes, which preserve meaning, from changes that alter what the text actually says. The pattern across all three is consistent. Less practiced writers change the smallest units or start again. More practiced writers change the structure and keep the object.
The second vocabulary comes from software, where starting over is expensive enough to have been argued about for fifty years. Brooks published the second-system effect in 1975 after watching engineers replace systems they had built before. The second design a person produces, he argued, is the most dangerous one they will ever make, because it begins from a list of everything they had to leave out of the first. Version two therefore arrives larger and more ambitious than version one, and takes longer to reach the point where anybody can judge it.
The third is about what the blank page genuinely buys, and this one grants the behavior something real. Luchins ran a set of water jar problems in 1942 and found that once people had learned a long method that worked, most kept applying it after a far shorter method became available, and many failed to see the shorter one at all. Jansson and Smith reported design fixation in 1991, showing that designers given an example solution reproduced its features in their own work, including features they had been explicitly told were defective. Once a solution sits in front of you, it is genuinely difficult to think past it. A fresh attempt breaks that grip, and that is not a rationalization, it is a measured effect.
The fourth names something specific about the moment you decide the current attempt is exhausted. Lucas and Nordgren have reported across several studies that people systematically underestimate how much they still have left in a creative task. Participants predicted their idea generation would fall off sharply after the first stretch, and it did not fall off nearly as far as predicted. The felt sense that the vein has run out arrives well before the vein has run out, and it arrives with the same confidence either way.
The fifth is where your version becomes unusual, and it is worth naming as such. The large measured literature on continuing against discarding runs almost entirely in the opposite direction from you. Sleesman and colleagues assembled 166 studies of escalation of commitment in 2012, and nearly all of that work is about people who kept investing in something that had stopped working. The failure the field is equipped for is over-persistence. Norton and colleagues found in 2012 that people place a higher value on things they assembled themselves than on identical things they did not, an effect they called the IKEA effect, which runs in the same direction: ordinary people over-attach to what they built. You are doing the reverse of both, which puts you somewhere the research has spent much less time.
Two honest cautions belong here, because both of these circulate in stronger forms than the evidence supports.
The first concerns the single most quoted sentence in favor of starting over. Brooks wrote in 1975 that you should plan to throw one away, because you will anyway, and that line has been used to justify a great many rebuilds since. Twenty years later he printed a retraction of it in the same book, arguing that the advice rested on a development model he had come to regard as wrong, and that building in increments beats building a first version intended for the bin. The most famous argument for scrapping work was withdrawn by the person who made it, and the retraction travels far less widely than the original.
The second concerns what can honestly be claimed about your direction of this behavior. The revision studies above are small, most of them followed student writers in the late nineteen seventies and early nineteen eighties, and none of them transfers cleanly to a spreadsheet, a kitchen renovation, or a product design. The escalation literature is large and points the other way. Nobody has assembled a body of evidence about the person who reliably destroys good work in order to remake it, which means anything you read that confidently states what such a person is doing, this page included, is reasoning past its data. What follows is a set of distinctions that hold up when people test them against their own week, offered as that rather than as findings.
Here is what all that vocabulary genuinely buys. It establishes that the pull toward the blank page has a documented mechanism behind it, that fixation is real and the fresh attempt does escape it, and that your sense of exhaustion mid-project is measurably premature. And here is where every one of those names stops. They describe the act. Two people can delete the identical draft on the identical evening and be doing entirely different things, one escaping a structure that really was wrong, one collecting the only three hours of the process that ever feel good. The vocabulary covers the what. The why sits outside all of it, and the why decides what would actually help.
About this page. This describes a psychological and behavioral pattern. It is not a diagnosis and it is not medical advice. I am not a therapist, and no page can tell you which of these roots is yours. Two people who recognize themselves in everything above can still need entirely different things. If you are in acute crisis, or if what you are facing involves your safety, your body, or a condition you are already being treated for, that belongs with a licensed professional now rather than later. I would rather help you find the right thing than have you keep reading here.
When Does This Belong With Somebody Qualified Rather Than a Page Like This?
Earlier than most people ask, and the line is drawn by what surrounds the restarting rather than by how many versions there are.
This page assumes two things about your situation, and both are worth stating plainly because everything after them depends on them. It assumes the restarting is costing you something real, and it assumes you are safe.
Read the next paragraph even if you are certain none of it applies. If the work you keep redoing is being redone because somebody else's reaction to a finished version would change the temperature of your house for a week. If you have destroyed work in order to avoid showing it to a specific person. If somebody else has destroyed your work, or controls whether it can be finished, and your income or your housing depends on not raising it. In any of those cases the redoing is doing safety work rather than craft work, and the two look identical from outside while needing completely different responses. Services for exactly that situation exist in most countries, many of them free and confidential, and the people staffing them speak with far more callers who are unsure than callers who are certain.
Two more that hide inside this pattern. If the redoing has spread well past work, so that ordinary completed actions get undone and performed again because they might not have been done correctly, and this costs you hours a day and stopping feels impossible rather than merely uncomfortable, that belongs with a licensed professional rather than with anything written here. It responds well to the right kind of help and poorly to willpower, and reaching for that help early is much cheaper than reaching for it after two years. If the destroying happens in episodes where you are sleeping very little, working at an unusual pace, and feel able to take on considerably more than usual, and those stretches are followed by long flat ones, that pattern also belongs with a professional and is worth raising even if you are unsure.
And if what accompanies this includes a flatness that has not lifted for months, sleep that has been broken for a long stretch, or any thought of harming yourself, that belongs with a licensed professional now rather than later.
The rest of this is written for the ordinary version, where the work is good, everybody thinks well of you, and you have quietly spent five months producing eleven versions of a thing that was usable in March.
Is Version Eleven Different From Version Two, or Only Later?
One test does nearly all the work on this page, it takes about five minutes, and you can run it tonight without telling anybody you ran it.
Open the version you were about to scrap and the version before it, side by side on one screen or printed on two sheets, and say out loud three differences that a stranger could confirm. Three real differences mean the restarts are producing a different result. One difference, visible only to you, means they are producing the same result at a later date.
That is the whole thing. Not three differences in how it feels. Three differences somebody who was not there could point at. A section that exists in one and not the other. A structure that runs in a different order. A decision that went the other way. Say them out loud, because saying them out loud is what stops the answer from being a general impression of improvement.
The second half of the test is the one that settles it, and it requires one other person and about four minutes of their time. Hand them both versions, unlabeled, and ask two questions. Which of these came first, and what changed. That reading is worth more than any amount of your own, for the reason set out above: they are reading the surface and you are reading the archaeology. If they cannot order them, and cannot name what changed, you have the cleanest possible evidence that the difference you are working so hard to produce is not reaching anybody.
| The restart is producing something | The restart is repeating | |
|---|---|---|
| Three differences, said out loud | Come easily, and they are structural | You get one, and it is a quality rather than a change |
| An outsider ordering the versions | Gets it right, and says why | Cannot order them, and asks what the difference is |
| What you can name as wrong | A specific decision, in one sentence | An overall feeling that it is off |
| Where the new version diverges | Early, at the decision you identified | Nowhere. It runs the same route |
| What the new version costs | Less than the flaw would have cost | More than fixing it, every time |
| Your best hours on the project | Spread across the whole of it | Only ever the first three of each version |
| Where you stop | Further along than last time | At the same place, on the same page |
Read the last row twice, because it is the cheapest signal here and it costs nothing to observe. If each version stalls at roughly the same point, the restart is delivering you back to the same obstacle with the earlier work removed rather than solving what stopped you.
One thing keeps this test honest. Plenty of people run it and find three genuine structural differences on every pair, and the correct conclusion in that case is that the redoing is working and this page does not describe them. The thing worth fixing then is that they are not keeping the versions, which is a filing problem and takes an afternoon.
Why Can Only You See the Seams?
Because you are the only person who watched them get made, and a seam that has been watched never becomes invisible again.
Take a wall with a patch in it. The person who lives in the room sees the patch every day for eleven years. Every visitor sees a wall. Neither is wrong. The resident's perception includes a history that the wall itself no longer contains, and no amount of looking at the wall will bring the visitor's version of it into reach.
Your nearly finished work is that wall, and the effect is stronger, because you did not merely watch the patching, you chose it. Faigley and Witte's distinction between surface changes and changes to what a text actually says is useful here in an unexpected way. The seams you cannot stop seeing are usually surface features carrying a memory of a structural compromise. The compromise was real. The visible residue of it, to anybody else, is a sentence that reads perfectly well.
There is a practical consequence and it is the most useful sentence in this section. Your judgment of a nearly finished thing you made runs harsh in a specific, measurable direction, and it runs harsher the more times you read it. That means one outside reading is worth more than four of your own, and it means the readings you already have are data. If three people have read it and none of them mentioned the thing you cannot stop seeing, that is not politeness. That is three independent measurements.
Before scrapping anything, write the seams down. All of them, as a list, in specific language. Most people find the list is between four and nine items, that three or four of them are twenty-minute fixes, and that two of them cannot be stated clearly enough to write down, which is itself informative.
Why Do Two People Scrap the Identical Draft for Completely Different Reasons?
Because the act of starting again carries no information about what produced it, and five very different needs resolve into the same blank page.
This is the limitation that makes general advice about finishing things so unreliable. Somebody watching you open a new file has exactly one observation available, which is the new file. They cannot see whether it came from an accurate structural judgment, from an inability to keep seeing the compromises, from the fact that beginnings are the only part that pays, from a doubt about the foundation, or from the plain preference for interesting work over dull work at nine in the evening. Neither can you, most of the time, because the judgment arrives as a perception rather than as a decision.
| Where it comes from | What the blank page buys you in the moment | The evidence in your own week |
|---|---|---|
| The old version holds the record of every compromise | A surface with no archaeology under it | You can list its flaws in detail and no reader has ever mentioned one |
| The clean slate is the reward | The only hours of this work that feel good | Your best sessions are always in the first three hours of a version |
| The foundation is doubted | Permission to stop building on something suspect | You stall shortly after the structure is set, not near the end |
| Repair is duller than building | Two interesting hours instead of two tedious ones | You restart when tired, late, or bored, almost never when fresh |
| The judgment was correct | An escape from something genuinely built wrong | The new version diverges early and other people can see it |
Only the last of those five is about the work. That is the whole reason a general policy of finishing what you start works so badly here. It aims at the blank page, which is identical in all five cases, rather than at what the blank page was for, which differs in every one.
The last column is available to you tonight without a conversation with anybody. Nothing there requires disclosure, apology, or a decision. It requires reading your own last three projects honestly for about fifteen minutes.
What If the Old Version Carries the Record of Every Compromise?
Then the restart is an attempt to stop seeing something, and the thing you want to stop seeing is not in the object.
This is the root that feels least like a pattern from inside, because it presents as accuracy. You are not imagining the compromises. Every one of them is real, you can date them, and if somebody asked you to defend the routing on page two you would not be able to. The reading is correct. What is doing the damage is that the reading is only available to you, and it gets refreshed every single time you open the file, so the object degrades in your hands while staying constant for everybody else.
The fixation research grants this root something the others do not get. Luchins and, later, Jansson and Smith showed that an existing solution genuinely does constrain what you can think of next, and that people reproduce features of what is in front of them even when warned that those features are defective. So the impulse toward a clean attempt has a real mechanism behind it and produces a real benefit. The blank page does buy you access to solutions the current draft is blocking.
What the research does not license is the destruction. Dow and Klemmer reported in 2010 that designers who developed several alternatives in parallel, and kept them, produced measurably better results than designers who worked serially and replaced each version with the next, and they were more willing to keep going. That is the move available here, and it costs nothing. Start the new attempt with the old one open in the next window rather than in a folder called old. You get the fresh route and you keep the comparison, which is the only thing that ever tells you whether the fresh route went anywhere.
The other move is the seam list from the section above, written before the new file is opened rather than after. A list of nine specific complaints is a work plan. A general sense that the thing is compromised is not, and it will regenerate identically inside the replacement, because the compromises were mostly forced by constraints that have not changed.
- Group
- 8–15 people
- Length
- Five days
- Where
- Sacred Valley, Peru
- You leave with
- Action steps
- The pattern
- It is doing a job
- The misread
- A character flaw
- The job
- Protection from something
- The tell
- It repeats on schedule
Five days in Peru's Sacred Valley, built around exactly the kind of pattern you just read about.
What If the Clean Slate Is the Reward?
Then the restart is the point of the activity, and finishing has been quietly reclassified as the thing that ends the fun.
Ask yourself where the good hours actually were on your last three projects. For a large number of people who recognize this page, the answer is the same on all three: the first afternoon. Nothing decided, every route still open, no compromise recorded, the whole thing still capable of being excellent. That stretch is genuinely enjoyable, and preferring it is a reasonable response to the fact that beginnings are made of better material than middles.
The difficulty is what happens when finishing pays nothing. Norton and colleagues found that the value people place on something they built themselves rises with the labor they put in, which is why ordinary people defend their own work past the point of sense. The condition attached to that finding is the important part here. The effect appeared when the building was completed and did not appear when it was interrupted before completion. Work that never reaches a finished state never earns the attachment that would otherwise make it hard to throw away, so each version is easier to scrap than the last, and the loop feeds itself.
This is where the accumulation problem sits, and it is the real cost of the pattern. All of that will only make a huge difference once you give it enough time and let time accumulate until it actually leads to something that is visible and palpable. Eleven versions amounts to the same first afternoon, eleven times, and it produces nothing visible or palpable at all, because nothing was allowed to stack on anything else. It also rarely is the first video that you post that gives you all of this visibility. The eleventh one is usually where that arrives, and only if the eleventh one follows the tenth rather than replacing it.
What moves this root is finding something at the end that pays. Not a general resolution to finish. One deliberately small thing, taken all the way to a state where somebody else sees it, chosen for being finishable in a week rather than for being important. What you are collecting is one instance of the ending producing a feeling, because you currently have very few of those and a great many instances of the beginning producing one.
What If You Do Not Trust the Foundation?
Then everything above it feels provisional, and provisional work cannot be finished at any price.
This root has a distinct shape and people recognize it by where the stalling happens. It does not happen near the end. It happens shortly after the structure gets set, at the moment the thing stops being an idea and starts being a commitment to one particular way of doing it. From there on, every hour of detail work carries a silent question about whether the detail is being applied to the right object. So the detail work does not get done properly, so the top layer stays rough, so the thing looks unfinished, which confirms the original doubt. The loop is tidy and completely airtight from inside.
Frost and colleagues measured a dimension in 1990 they named doubts about actions, a persistent uncertainty about whether something has been carried out properly, and it sits at the center of this root rather than across the other four. The important detail is that the doubt attaches to the action rather than to the result, which is why more evidence of the result being fine does nothing to settle it.
The move here is narrow and it is the opposite of rebuilding. Write down the structural claim you doubt, as one sentence, in terms specific enough that it could be checked. This section runs in the wrong order. This model breaks if the input is a range rather than a number. The layout will fail on a narrow screen. Then test that one claim directly, on the existing version, in an afternoon. Two things happen. Sometimes the claim is true, in which case you have found the actual decision that was wrong and a restart may well be correct. More often the claim cannot be written down at all, and the inability to write it is the finding. A doubt that cannot be stated as a claim points at the experience of building on something you did not get to see finished.
What If Repair Is Just Duller Than Building?
Then you have the most tractable of the five, and it responds to scheduling rather than to insight.
Grant the premise first, because most advice on this skips it and loses the reader. Repair is duller. This is not a distortion you need to correct. Finding out why the totals are off by a small amount means reading your own formulas backward through eleven columns while holding the whole model in your head. Moving a wall that is already up is worse work than deciding where to put it. Revising a chapter means reading your own sentences closely enough to find the one that is doing the damage, which is the least enjoyable reading there is. Building is generative and repair is diagnostic, and the two use different equipment.
The revision research lands exactly here. Sommers found that the less practiced writers had no procedure for locating what was wrong, so their revision collapsed into changing words or starting over, while the experienced ones had a way of naming the dissonance and working on it. The skill that separates them is a diagnostic one, and it is learnable, and almost nobody is taught it.
The tell for this root is timing rather than content. You restart at night, at the end of a week, when tired, when bored, after an interruption. You almost never restart at nine on a Tuesday morning with a clear head. If you graphed your restarts against the hour of the day you would have your answer inside ten minutes.
The fix is unglamorous and it works. Put repair in your best hours instead of your leftover ones, which usually means the first ninety minutes of a working day rather than the last ninety of an evening. Time-box it, because diagnostic work with no end point is unbearable and diagnostic work with a bell on it is merely dull. And define what you are looking for before you start, so the session has a target rather than a mood.
What If the Thing Genuinely Was Wrong?
Then you were right, and this deserves the first hearing rather than the last one.
Sometimes a piece of work is built on a decision that cannot be repaired into a different decision. A structure chosen in week one determines what can be built in week six, and if week one was wrong, week six is spent decorating something that should not exist. Craftspeople redo work constantly, and the redoing is frequently the reason the work is any good. Nothing on this page argues for accepting worse output, and any advice that ends at settle for it has misread the problem.
The check has three parts and all of them happen before the new file opens.
First, name the decision that was wrong, in one sentence, at the level of a decision rather than a quality. The sequence starts in the wrong place. This is built around a monthly cycle and the real cycle is weekly. The main character wants the wrong thing. If the sentence will not come, what you have is a verdict on the object rather than a fault in it, and a replacement will inherit whatever produced the verdict.
Second, say what the new version will do differently at that exact point, before you start it. Write it at the top of the new file. When the new version reaches that point, read what you wrote and check.
Third, budget it. A two-hour thing redone costs two hours and the arithmetic barely matters. A five-month thing redone is a decision with a real price on it and deserves an hour of thought and a written reason, which is roughly one percent of what it will cost you.
Brooks's warning applies from here onward. The rebuild starts from a list of everything you resented about the first version, none of the original constraints come with it, and version two therefore arrives larger and more ambitious than version one. It also takes longer to reach the point where it can be judged, which is exactly the point you need it to reach before your interest runs out. Cap the scope of the new version at the scope of the old one, and add nothing that was not on your list of faults.
And it would be humorous to think that you can just make a plan that's going to work on the first strike. Life rarely works that way. The thing that separates useful redoing from a loop is the second half of trial and error, which is knowing how to evaluate what came back. Most people running this pattern do a great deal of trial and almost no evaluation, because the evidence gets filed in a folder called old and never opened again.
What Is the Difference Between Setting Something Aside and Destroying It?
One buys you distance and keeps the option. The other buys you the same distance and removes it.
These two get confused constantly, and the reason is that both produce relief. The relief of putting something in a drawer for two weeks and the relief of starting again feel similar in the hour they happen, and only one of them can be reversed.
Sio and Ormerod reviewed the incubation literature in 2009 and found a real effect from setting a problem aside, larger for open-ended problems than for tightly defined ones. The honest version of that finding is modest. The effects are small, the conditions that produce them are not stable across studies, and nobody should treat two weeks in a drawer as a guaranteed fix. What it does reliably provide is the only thing that reduces your archaeology problem, which is time between readings.
The practical version is a drawer with a date on it. Put the thing away, write the date you will next open it, and do not open it before then. Two weeks is usually enough. When you do open it, read it once, from the top, without touching anything, and only then decide. A large number of people find at that point that the version they were certain was unsalvageable is obviously fixable, and that the flaw that ruined it is one paragraph.
If you have already deleted something recently, look for it before accepting that it is gone. Version history, trash, autosave folders, the email you sent yourself, the message where you pasted a section to somebody for a second opinion. Recovered drafts are almost always better than the memory the deleter had of them.
Where Does This Hide When You Would Not Call It Starting Over?
In about seven places nobody would file under restarting, which is why most people undercount this by a wide margin.
The most common is the thinking document. You open a new file just to work something out, with no intention of it becoming the real one, and four hours later it is the real one and the previous version has been retired without a decision ever being made.
Then there is the rename. A file called final-v3-new is a restart wearing a filename, and a folder with six of those in it is a record of six restarts nobody counted, including you.
Rebuilding rather than debugging belongs here in every medium. The spreadsheet rebuilt from source data instead of the formula being found. The document reformatted from scratch instead of the style being fixed. The room stripped instead of the section being repaired.
The reread from the top is the quietest one and it may be the most expensive. Every session begins at page one, with editing, so the first third of the work gets forty passes and the last third gets none. Nothing has been deleted and the effect is identical to a restart, because the end never gets built.
Small physical versions count too and they add up. Retaking the photograph. Re-recording the voice message. Redoing the run because the watch did not start. Rewriting the email that had already been fine, which has its own mechanism and its own page.
Then the life-sized version, where the job, the city or the relationship gets replaced rather than repaired, and the new one develops the same faults on a similar schedule. That one is a much bigger subject and it has a page of its own.
And the plan-shaped version, where nothing physical is destroyed at all but the schedule gets rewritten from a blank page every Sunday, so no week ever builds on the one before it. The broken week and its reset are covered separately.
Is "It Was Not Good Enough" an Explanation or an Excuse?
Both, in different weeks, and the split between them is testable in about a minute.
An explanation names a specific fault and predicts what the replacement will do differently. The sequence starts in the wrong place, so the new one will open at the decision rather than at the background. That sentence can be checked later, and it will either turn out to be true or it will not, which is what makes it an explanation.
An excuse names a quality and predicts nothing. It was weak. It was not right. It did not have it. Those sentences cannot be wrong, which is precisely what makes them useless, and they will describe version twelve exactly as well as they described version eleven.
The reason this distinction matters is that the second kind is unfalsifiable and therefore permanent. A standard you cannot state in advance gets set by whatever object is currently in front of you, which means the object always falls short of it, because it is generating the standard it is being measured against.
Simon described satisficing in 1956 as searching until you find something that meets a standard you already hold, rather than searching for the best available option. The operative part is that the standard exists before the search. Write down, before you begin, what this thing has to do for it to be finished. Three or four specific items. Then measure the finished version against that list rather than against a feeling. Most people who try this discover their aspiration level has never once been written down in their working life, and that when it is, the current version meets it.
There is a related distinction worth keeping. An explanation of where something came from is genuinely useful, right up until it becomes the account you give yourself for not facing the thing. The test is whether it changes what you do next.
What Does the Restarting Actually Cost You?
More than the hours, and the hours are the part that is easy to count.
The time is real and it is the smallest item. Five months of eleven versions where three months would have produced one finished thing plus two revisions is four wasted months, and most people already know that number roughly.
The accumulation cost is larger and almost invisible. Nothing compounds. Every project restarts at the same distance from the finish line, so the distance never gets shorter. It takes time and repetition to build something that is of significance, and repetition means the same thing continuing, not the same beginning recurring.
The skill cost is the one people are most surprised by, and it explains something they have often noticed without connecting it. You have thousands of repetitions of starting and very few of finishing. So the skills that live at the end of a project, which are diagnosis, revision, deciding a thing is done, and tolerating an object that has visible flaws in it and letting it out anyway, have almost no practice behind them. They feel hard because they are undeveloped rather than because you are unsuited to them, and undeveloped skills develop.
The evidence cost is quiet. You never find out whether the flaw you fixated on mattered, because no version carrying it ever reached anybody. So the question stays open, and an open question generates more restarts. Fifteen years of this leaves you with no calibration at all about which of your standards are load-bearing.
And there is a slow professional cost that nobody announces. What people count is finished work, because versions are invisible outside your own machine. Colleagues keep rating your ability highly, because your beginnings are genuinely good, and quietly stop routing the long items toward you, because they have not seen many endings. Most people notice this years after it opens and do not connect it to a folder of eleven files.
One thing worth saying plainly, because it is the fear that stops people looking at this at all. None of that makes you a person who cannot finish things. This behavior clusters in people with real standards, and the standards are frequently the reason their work is worth anything. The cost is real and the character reading is wrong, and both of those can be held at once.
Why Does the Rebuilt Version Come Out Bigger Rather Than Cleaner?
Because it starts from a list of grievances rather than from a scope, and grievances have no ceiling.
Watch what actually gets carried into version two. Every compromise you resented in version one gets promoted into a requirement. The section you had to cut goes back in. The structure you settled for gets replaced by the one you wanted. The feature you left out because there was no time is now in the plan, because this time you are doing it properly. Meanwhile the constraints that forced each of those compromises, which were real, do not come along, because constraints live in the situation rather than in the file.
Brooks watched this happen to engineers who had built one system and were building its replacement, and the pattern he described is exactly the shape above. The second system attracts every refinement withheld from the first, which makes it slower to reach a working state.
The practical consequence is specific and it is the reason so many rebuilds die younger than the thing they replaced. A bigger version takes longer to reach the point where it can be judged, and the point where it can be judged is where your interest historically runs out. So version two is more likely than version one to be abandoned before it gets there, and version three is drafted from an even longer list.
Two rules stop it. Write the scope of the new version before the first line, and cap it at the scope of the old version. Add nothing that was not on your written list of faults, and put every improvement you think of during the rebuild into a separate list for later, where it can wait without slowing anything down.
Why Does Deciding to Just Finish It Not Work?
Because it aims at the wrong layer, at the wrong moment, and it treats a five-cause pattern as though it had one cause.
Take the moment first. The decision to scrap does not present itself as a decision at all. It arrives as a perception that the work is wrong and that starting again would be faster, and perceptions do not feel like something you are choosing. A general resolution to finish what you start lives in deliberate thought, so it will be present and effective in every project where you were never going to restart, and absent in the one where the document goes gray at eleven at night.
Then the aim. Deciding to finish addresses the blank page, which is identical across all five roots and therefore carries no information about any of them. For the tiredness root it is roughly the right target and still fails on timing. For the other four it is pointed at the wrong object.
There is a third reason and it is the one that catches thoughtful people. Understanding which root is yours changes nothing on its own, because the required action differs in each case and none of them is finishing harder. For the archaeology root it is keeping the old version open beside the new one. For the clean slate root it is one small finished thing that produces a feeling. For the doubted foundation root it is a single structural claim, written down and tested directly. For the tiredness root it is repair scheduled into your best hours. For the correct judgment root it is a written sentence naming the decision that was wrong, checked when the new version reaches it. The general problem of understanding something and continuing to do it anyway is covered separately.
Expect something specific when you stop, because it gets misread as failure. Keeping a flawed version and repairing it will feel flat. Something that was reliably producing a small, clean, immediate good feeling has been removed and nothing has replaced it yet, and the gap lasts a few weeks. That is the ordinary shape of removing anything that was quietly doing a job, and it closes.
What Would You Have to Give Up to Keep the Flawed Version?
You have to give up more than you would expect to give up, and naming it in advance is what stops the change collapsing in week three.
You give up the relief. The clean page arrives reliably, it costs nothing, and nothing else in your working week supplies that particular feeling.
You give up the first three hours, repeatedly. They are the best hours the work contains and you have been collecting them at a rate of one set per version. Keeping a version means fewer of them and more of the other kind.
You give up a description of yourself. If you have been the person with standards high enough that they would never let something compromised out of the building, then finishing a thing with three visible flaws in it retires part of that description, and the adjustment is real rather than trivial.
And you give up a specific private hope, which is the hope that the next one will finally come out right. Every new file carries it. Finishing a flawed version means settling that question for this project, in the direction of not quite, which is the exact discomfort the restarts have been postponing.
What arrives instead takes longer to notice and is worth considerably more. The eleventh version becomes the eleventh rather than the first again. An object exists that other people can respond to, which produces information you cannot obtain any other way. And after two or three finished, imperfect things, you find out which of your standards actually mattered to anybody, which is the piece of calibration this pattern has been keeping from you for years.
What Does Repair Look Like on an Ordinary Tuesday?
Duller than a restart, shorter than you expect, and with a written list in front of it.
| How it runs now | Where it goes | What that sounds like on a Tuesday |
|---|---|---|
| The document goes gray and a new file opens | The seams get written down as a list first | Nine items. Four are twenty-minute fixes |
| The old version goes into a folder called old | The old version stays open in the next window | You check the new one against it at the halfway point |
| A structural doubt gets rebuilt around | The doubt gets written as one testable claim | This breaks if the input is a range. Check it by lunchtime |
| Repair happens at eleven at night, or never | Repair takes the first ninety minutes of the day | Timer on, one target, stop when it rings |
| Good enough is decided by looking at the object | Good enough is three written items, set in advance | It does these three things, so it is done |
| Eleven versions nobody can tell apart | Three versions somebody could order | You hand two to a colleague and they get it right |
| The best hours are the first three of each version | The best hours are spread across one thing | Week five is still interesting, because it is still the same thing |
If the middle column reads as out of reach tonight, hold off on any verdict about yourself. The left column was built by a real person doing real work under real pressure, every entry in it was solving something at the time, and most of what it was solving was reasonable.
Are You Sure That Is What You Are Looking At?
Six behaviors get described in nearly the same words, and only one of them is the pattern on this page.
| What you are describing | What it probably is | Where it is covered |
|---|---|---|
| Work that reaches the edge of finished and stops, intact | Non-delivery at the end of the arc | Abandoning things at ninety percent |
| Nothing ever begins because there is more to read first | A preparation loop | Researching instead of starting |
| The plan gets rewritten every Sunday and dies on Wednesday | A future start date and a weekly reset | Starting again on Monday |
| Four lines of a message taking thirty-five minutes | Editing an outgoing communication | Rewriting messages |
| A standard nothing ever meets, across everything | The wider standard itself | How to stop being a perfectionist |
| A job, a city or a life replaced rather than repaired | The same move at life scale | How to start over |
| Completed work destroyed so it can be made again | This page |
Two of those rows deserve an extra line, because the overlap is genuine rather than superficial. If you both abandon things near the end and restart the ones you keep, you are running two patterns in sequence, and the earlier one is the restart, so that is where the work is. And if the standard is the whole issue, applying to your cooking and your emails and your appearance rather than to your projects, this page will describe your Tuesdays accurately and fix less than the standard page will.
The distinguishing feature here is worth naming plainly. Something usable existed. The question is why it was removed rather than repaired.
Which Root Is Yours, From Evidence You Already Have Tonight?
From seven questions you can answer alone, tonight, without speaking to anybody.
- Run the two-version test. Open the last two versions and say three differences a stranger could confirm. Three structural differences means the restarts are working. One quality-shaped difference means they are repeating.
- Find where each version stopped. Note the page, the section, the stage. If all of them stalled in the same place, the restart is delivering you back to the same obstacle.
- Check the clock. When do you restart? If it clusters at night, at the end of weeks, and after interruptions, the tiredness root is live and it is the most tractable one.
- Check the reader evidence. Has anybody who read your work ever named the flaw you cannot stop seeing? If several have read it and none has mentioned it, you have independent measurements and you are ignoring them.
- Try to write the fault as a sentence. One sentence, at the level of a decision. If it comes easily, your judgment may well be correct. If it will not come, the verdict is about the object rather than about a fault in it.
- Ask where the good hours were. If every project's best sessions were in its first afternoon, the beginning is paying you and the ending is not, and that is a different problem from a quality problem.
- Ask when the stalling started. Right after the structure got set, or near the end? Early stalling points at the doubted foundation. Late stalling points somewhere else, and probably at the ninety percent page rather than this one.
Two or three of these will agree with each other, and where they agree is your root. Where they disagree you probably have two, which is common and not a complication. Most people find one dominant root plus a second that shows up in one specific kind of work.
It is good to have one's blind spots uncovered, and this particular uncovering can be done privately and cheaply before anybody else is involved at all.
- Group
- 8–15 people
- Length
- Five days
- Where
- Sacred Valley, Peru
- You leave with
- Action steps
- The premise
- No behavior is random
- The work
- Find the cause underneath
- The change
- Built on your calendar
- The point
- Life changes through action
If reading about it is no longer enough, the seminar is where the pattern gets worked on.
How Long Does This Take to Change, Honestly?
Faster than the guilt version suggests and slower than the resolution version does, and it runs at five different speeds.
Noticing arrives in about two weeks with a written log of every version you make and every one you retire, including the renames and the thinking documents. That stage is uncomfortable in a specific way, because you will catch yourself opening the new file and open it anyway. Seeing it counts, and everything after it depends on it.
The tiredness root moves fastest, often within days, because it needs a scheduling change rather than an insight. Repair moved into your first ninety minutes, with a timer and a target, changes the ratio almost immediately.
The archaeology root moves within a few weeks and it moves on one habit, which is keeping the old version open beside the new one instead of filing it. It feels like clutter for about five attempts and then it starts producing comparisons you can actually use.
The doubted foundation root moves when one structural claim gets written down and tested directly, which can happen in an afternoon. What takes longer is trusting the result, because the doubt reattaches to the next structure and has to be tested again two or three times before it loses its authority.
The clean slate root takes longest, because a finished thing has to start producing a feeling, and that requires several finished things. Expect a few months, and expect the first one to feel anticlimactic, which is normal and is not a verdict on the work.
Where Does This Thinking Come From?
From public bodies of work, none of which require going through me to check.
On revision, Sommers published the study of student and experienced adult writers in 1980 in College Composition and Communication. Perl published the work on the composing processes of unskilled writers in 1979 in Research in the Teaching of English. Faigley and Witte published the surface and meaning revision taxonomy in 1981, also in College Composition and Communication. All three are small studies of writers, mostly students, from that period, and their transfer to other kinds of making is an inference rather than a finding.
On rebuilding, Brooks published The Mythical Man-Month in 1975, which contains both the second-system effect and the line about planning to throw one away, and the 1995 anniversary edition contains his retraction of the second of those.
On what a fresh attempt buys, Luchins published the water jar experiments on mental set in 1942, and Jansson and Smith published the design fixation studies in 1991 in Design Studies. On how much is left in a task when it feels exhausted, Lucas and Nordgren have published a series of studies on the underestimation of creative persistence from 2015 onward.
On value and completion, Norton and colleagues published the IKEA effect in 2012 in the Journal of Consumer Psychology, including the finding that the effect depends on the building being completed. On the other direction, Sleesman and colleagues published the meta-analysis of 166 escalation of commitment studies in 2012 in the Academy of Management Journal.
On keeping alternatives rather than replacing them, Dow and Klemmer published the parallel prototyping work in 2010 in ACM Transactions on Computer-Human Interaction. On setting things aside, Sio and Ormerod published the meta-analytic review of incubation in 2009 in Psychological Bulletin, and the effects there are real and small. On doubts about actions, Frost and colleagues published the dimensions of perfectionism in 1990 in Cognitive Therapy and Research. On stopping rules, Simon published the satisficing account in 1956 in Psychological Review.
The rest is mine and carries no research weight at all. I hold no clinical qualification of any kind, and I am not a therapist. None of the researchers named here has any connection to this page or to the seminar. Where their published work and my own reading disagree, theirs is the one to go and check.
What Is Worth Doing Before the Next Time You Reach for a New File?
Seven things, and the first four ask you to change nothing about how you work.
- Log versions for two weeks. Every new file, every rename, every thinking document that became the real one. Change none of it. You are collecting a baseline, and a baseline from a normal two weeks is worth more than one from a good week.
- Mark where each version stopped. Page, section, stage. This single column tells you within two weeks whether your restarts are moving you forward or returning you to the same obstacle.
- Run the two-version test on your last pair. Side by side, three differences said out loud. Five minutes, tonight, and nobody needs to know you did it.
- Get one outside reading. Hand two versions to somebody who has not seen either and ask which came first and what changed. Four minutes of their time, and the most useful information available to you on this whole subject.
- Write the seams down before you scrap anything. All of them, specifically. Then count how many are twenty-minute fixes and how many cannot be written down at all.
- Stop deleting and start keeping. Old version open in the next window, new attempt beside it. This costs nothing, removes no options, and changes the behavior more than any resolution about finishing does.
- Finish one small thing badly. Chosen for being finishable in a week rather than for being important, and taken all the way to a state where somebody else sees it, with its flaws in place. You are collecting one instance of an ending producing a feeling.
That sequence costs almost nothing and produces something no article can hand you, which is a measurement of your own loop in your own handwriting. Then put the repair sessions into your real calendar and hold the calendar, because a decision that lives only in your head has no way of surviving eleven at night.
If what landed hardest was the two-version test, and specifically the possibility that nobody else could tell your versions apart, our Sacred Valley seminar works at the layer underneath the versions rather than at the level of the work. What happens there is that we find what the blank page has been doing for you, since that is the part no productivity method reaches, and then build something that does the same job without costing you the object. I cannot promise you that five days will make you a person who ships version two. I can certainly set you in the right direction, and I would rather say that than a better sentence I could not stand behind. Everybody is screened beforehand, which keeps the group small enough that somebody who has spent fifteen years producing excellent beginnings cannot quietly keep doing it inside the room. The weeks afterward are part of the design rather than something left to you, because the real work only starts when you come home to the same file, the same folder called old, and the same eleven at night. One limitation worth stating plainly: exact dates, the price and the final cap on group size are still being settled, so if you are comparing programs on those numbers this week, ours cannot yet be compared. Ask every program you are weighing the same questions, including that one, and decide from your own reading of what comes back. The two-week log costs you nothing and is worth having whatever you decide afterward.
Questions About Scrapping Work and Starting Again
Why do I scrap good work and start again from nothing?
Because the version in front of you holds the record of every compromise that went into it, and you are the only person who can see that record. A blank page clears all of it in one keystroke, and the relief arrives immediately. Sometimes the judgment behind the scrapping is correct and the work really was built wrong. The useful question is whether anyone but you could tell your versions apart.
Is starting over always a mistake?
No, and any page that says so is asking you to accept worse work. People who redo things are often the reason the work is good at all. The distinction that holds up is whether the redo answers a question the last version raised, or whether it repeats the last version's first three hours. One produces a different result. The other produces the same result at a later date.
What is the difference between revising and restarting?
Revising keeps the object and changes it. Restarting removes the object and produces another one. Sommers described the split in 1980 after interviewing student writers and experienced writers, and found the two groups meant different things by the same word. Students talked about revision as cleaning up and doing over. Experienced writers talked about finding the shape of the argument inside what they already had.
Why does a blank page feel better than a nearly finished one?
Because a blank page carries no history. Everything you have already made holds the trail of the decisions that produced it, including the compromise on page two and the part you could not get working so you routed around. A new file has none of that yet. It is flawless for the simple reason that it is empty, and that flawlessness lasts about twenty minutes.
How do I tell a restart that helps from one that repeats?
Open the version you are about to scrap and the version before it, side by side, and say out loud three differences a stranger could confirm. Three real differences mean the restarts are producing something new. If you can only name one, and only you can see it, the restarts are producing the same result on a later date. The test takes about five minutes.
Why can only I see the flaws in my own work?
Because you were present for every decision and the losing options stayed in your memory attached to the object. The reader sees a wall. You see the patch, because you watched it get patched. That asymmetry never resolves on its own, which is why an outside reading is worth more than another hour of your own, and why your own judgment of a nearly finished thing runs consistently harsh.
What is the second-system effect?
Brooks named it in 1975 after watching engineers build replacements for systems they had made before. The second design a person produces tends to be the most dangerous one, because it starts from a list of everything held back the first time. Version two therefore arrives bigger and slower than version one, takes longer to reach the point where you can judge it, and gets scrapped before it does.
Is it true that you should plan to throw the first one away?
That line comes from Brooks in 1975 and it is the sentence most often quoted in defense of restarting. Twenty years later he printed a retraction of it, arguing that the advice rested on a development model he had come to think was wrong, and that building in increments beats building a version intended for the bin. The most famous argument for scrapping work was withdrawn by the person who made it.
What is design fixation?
Jansson and Smith reported in 1991 that designers shown an example solution reproduced features of it in their own work, including features they had been told were defective. Luchins had shown something similar in 1942 with a set of water jar problems, where people kept applying a long method after a much shorter one became available. Once a solution is in front of you, it is genuinely hard to see past.
Why are my best hours always at the start?
Because the opening hours of anything are made of a different material. Nothing is decided yet, every option is still live, and no compromise has been recorded. That stretch is genuinely enjoyable, and preferring it is not a defect. The difficulty starts when the enjoyment becomes the reason for the restart, because then the beginning has stopped being a beginning and become the whole activity.
What is the IKEA effect?
Norton and colleagues published it in 2012, showing that people place a higher value on things they assembled themselves than on identical things they did not. The part that matters here is the condition attached. The effect showed up only when the building was completed. Work that never reaches a finished state never earns the attachment that would otherwise make it hard to throw away.
Why do I rebuild the spreadsheet instead of fixing the formula?
Because finding the formula means reading your own work closely, backward, in the place you least want to look, while rebuilding means an hour of pleasant construction with a clear finish line. Rebuilding is also faster in the first twenty minutes and slower over the whole afternoon, and the first twenty minutes are what you are estimating from when you decide.
Is this perfectionism?
It overlaps and it is narrower. Perfectionism describes a standard that stays out of reach. This page describes a specific action taken because of one, which is the destruction of usable work in order to begin again. Frost and colleagues measured a dimension in 1990 they called doubts about actions, a persistent uncertainty about whether something has been done properly, and that dimension maps onto one of the five roots here rather than all of them.
Why is my behavior the opposite of the sunk cost effect?
Because the measured literature is nearly all about people continuing too long. Sleesman and colleagues pulled together 166 studies of escalation of commitment in 2012, and the failure the research equips you for is over-persistence. Running it backward, and discarding investment other people would defend, is far less studied. That makes your version genuinely interesting and it also means less confident advice exists about it.
Why do I restart most when I am tired?
Because repair asks for the sharper attention and building asks for less. Tracking down what is actually wrong with something requires holding the whole object in your head while reading it against itself. Producing a new opening requires none of that. At eleven at night the new file is available and the repair work is not, so the decision gets made by the hour rather than by the work.
How many versions is too many?
There is no number, and any number would be wrong for somebody. The measure that works is whether the versions are distinguishable. Eleven versions that another person can sort into an order and describe the differences between is a person iterating. Eleven versions that nobody can tell apart is the same afternoon, repeated eleven times, at a cost of eleven afternoons.
Should I keep the old versions?
Yes, and keeping them changes the behavior more than any resolution about it does. Dow and Klemmer reported in 2010 that designers who developed several alternatives in parallel and kept them produced better results than those who replaced each version with the next. Keeping the old one also converts every future restart into a comparison you can actually run rather than a fresh start you cannot check.
What does good enough actually mean?
In practice it means a threshold you set before you look at the object, which is the part most people skip. Simon described satisficing in 1956 as searching until something meets a standard you already hold rather than searching for the best available option. A standard set after the fact, by whatever is in front of you, can never be met, because the object supplies the standard and always falls short of it.
Why does the new version end up bigger than the old one?
Because it starts from a list of grievances rather than from a scope. Every compromise you resented in version one gets promoted into a requirement for version two, and none of the original constraints come along to hold them down. The result takes longer to reach a judgeable state, which is exactly the state you need it to reach before your interest runs out.
Am I just procrastinating?
Sometimes, and the tell is where the work stops. Procrastination usually means the work does not begin. This behavior involves substantial real work, done well, and then removed. It can serve the same purpose as delay while looking like effort from the outside and feeling like effort from the inside, which is why it survives so much longer than ordinary avoidance does.
What if I was genuinely right that it was wrong?
Then you were right, and this happens often enough that it deserves the first hearing rather than the last. The check is whether you can name the specific decision that was wrong, in one sentence, before you start the replacement, and whether the replacement makes a different decision at that exact point. If you cannot say what was wrong at the level of a decision, the judgment was about the object rather than about a fault in it.
How is this different from abandoning things at ninety percent?
Abandoning leaves the thing intact and unfinished, so it sits in a folder and could still be picked up. This behavior removes the thing and starts another one, so the pile is versions rather than fragments and the activity looks productive from every angle. Anyone who both abandons and restarts is running two patterns, and the abandoning page covers the stop near the end.
How is this different from never starting at all?
Research loops and preparation loops keep the work in the future, so nothing exists to look at. This page assumes something already exists and got destroyed, which requires having done the work first. The two can alternate in the same person, usually with the research phase reappearing right after each restart, and the never-beginning version has its own page and its own roots.
Does setting work aside help, or is it the same avoidance?
It helps, modestly, and it is a different action from destroying. Sio and Ormerod reviewed the incubation literature in 2009 and found a real but small effect, larger for open-ended problems than for closed ones. A drawer gets you the distance that makes rereading possible while keeping the option of returning. Deleting gets you the distance and removes the option in the same movement.
Why does my colleague ship at version two?
Usually because they cannot see what you can see. They were present for fewer decisions, they hold no memory of the version where the compromise was not yet a compromise, and their standard is set by what the thing has to do rather than by what it could have been. Their version two and your version eleven are frequently indistinguishable to everybody else.
What do I do with the draft I already deleted?
Look for it before you accept that it is gone, because it usually is not. Check the version history, the trash, the autosave folder, the email you sent yourself, the message where you pasted a paragraph to somebody. Then read it cold, from the top, without touching it. Most people find it is markedly better than the memory they had of it at the moment they deleted it.
Does this happen more at work than at home?
It varies by person and the pattern in the answer is useful. If it happens only where the work will be assessed, the restarting is doing something about how you will be judged. If it happens on private projects nobody will ever see, that reading is unavailable and the cause sits somewhere else, usually in the pleasure of beginning or in a doubt about the structure.
Why does deciding to just finish it not work?
Because the decision to scrap does not arrive as a decision. It arrives as a perception that the thing is wrong and that starting again would be faster, and it feels like clarity rather than like a choice. A general resolution to finish things lives in deliberate thought and will be reliably available in every situation except the one where the perception shows up.
What is the fastest thing I can change this week?
Stop deleting. Rename instead, put the old version in a folder you keep, and start the new one beside it rather than on top of it. That single change costs nothing, removes no options, and converts your next restart into a comparison you can check. Most people find the comparison does more than any resolution they have made about finishing things.
How long does this take to change?
Noticing takes about two weeks with a written log of every version you make and every one you scrap. The tiredness root moves within days once repair gets scheduled into your best hours instead of your leftovers. The clean slate root takes a few months, because a finished thing has to start producing a feeling. The doubted foundation root moves when one structural doubt gets tested directly rather than rebuilt around.
Does this hold a career back?
Quietly, and rarely in a way anybody names out loud. What erodes is the record of finished work, since versions do not count and only shipped things do. People stop routing the long items toward you while continuing to rate your ability highly, because they have watched you produce excellent beginnings and have not seen many endings. That gap is usually noticed years after it opens.
Want the part most retreats skip?
Join the list for the next Sacred Valley seminar: integration-first deep inner work, an intentionally small group, and tools you actually take home.
See the seminar florian@florianrolke.com