Pick a fault you’ve tried to fix.
Not a habit — a fault. Something about how you operate that you’ve decided is a problem. The interrupting. The sharpness when you’re tired. The way you go cold instead of saying the thing. The drinking a bit more than you’d like.
You’ve attempted it before. Possibly several times.
And what happened was: a period of success, then a bad week, then it was back — and when it came back it didn’t feel like a relapse. It felt like the truth reasserting itself.
Here’s the thing almost nobody tells you.
There are three different places you can attack a fault, and they’re not equally effective. Most people only ever attempt the first one.
That’s why nothing has held. Not insufficient willpower. Wrong level.
The Three Levels
A fault can be corrected at three levels, which differ in durability. Correction at the event addresses the behaviour itself, through vigilance, restriction, or suppression of the specific action. Correction at the reasoning addresses the belief supporting the behaviour, by establishing why it is mistaken so that the argument for it no longer holds. Correction at the source addresses whatever generates the impulse — the underlying disposition, appetite or self-concept from which the behaviour follows. Event-level correction fails under pressure, because it requires continuous effort and the behaviour reappears whenever attention or resource is disrupted. Reasoning-level correction is more durable and generalises across situations, but fails when deliberate reasoning is unavailable, which is precisely when the behaviour tends to occur. Source-level correction holds without maintenance, because the impulse is no longer produced, but it is slower and produces no visible output for an extended period.
This is not a new framework. It’s roughly four hundred years old, and it comes from the same source as everything else on this site — the second of Liaofan’s four lessons, written by a man who’d audited his own character in writing and needed a method for actually changing what he found.
Let me run one fault through all three so you can see the difference.
The Same Fault, Three Ways
The fault: you interrupt people. Specifically, when someone is taking too long to reach their point, you finish it for them. You’ve been told about it. You know it lands badly. You’ve decided to stop.
Level one: correct at the event
You resolve not to interrupt.
In practice this means monitoring. You watch yourself in conversations, catch the impulse, and hold it. It requires attention, and it works — for a while, and specifically in the meetings where you remembered to be watchful.
Then there’s a day where you’re behind, under pressure, three conversations back to back, and somebody takes ninety seconds to make a twenty-second point.
You interrupt. Twice.
Why it fails: you never touched the impulse. You built a filter downstream of it and staffed the filter with attention, and attention is exactly what’s unavailable under pressure. The moment the resource is disrupted, the output resumes.
Worse: this level produces a specific and demoralising pattern. Success in easy conditions, failure in hard ones — which reads as evidence that you can do it when you try, and therefore that failing means you didn’t try. That reading is wrong and it’s the source of an enormous amount of unnecessary self-criticism.
What it’s good for: containment. It genuinely stops the behaviour in bounded situations, and that’s not nothing. It’s just not change.
Level two: correct at the reasoning
You go a level down and ask why you do it.
And you find an argument underneath: I’m saving time. I already know where this is going. Finishing their sentence is efficient.
Then you examine it properly, and it doesn’t survive. You’re right about where they’re going maybe two-thirds of the time, and wrong the rest — and the times you’re wrong are expensive, because you’ve now responded to a point they weren’t making. You’ve also trained several people to give you less information, which costs you far more than the seconds you saved.
The argument is defeated. And this is a real improvement over level one, because it generalises — it applies in every conversation, not just the ones where you’re being vigilant.
Then you’re tired, at the end of a long day, and it happens anyway.
Why it fails: the behaviour isn’t produced by the argument. The argument was constructed afterwards, to justify something already happening. Defeating a justification doesn’t remove the thing it was justifying, and reasoning is unavailable at exactly the moments the impulse is strongest.
This is the level most thoughtful people stall at. They’ve done genuine intellectual work, they’ve reached an accurate conclusion, and the behaviour is unchanged — which is the insight-action gap in its most frustrating form.
What it’s good for: it stops you defending the behaviour, which is a genuine prerequisite. You can’t work on something you’re still justifying.
Level three: correct at the source
Now the question changes. Not “how do I stop” and not “why is it wrong,” but:
What in me produces the impulse in the first place?
For the interrupting, it usually isn’t impatience with time. Underneath there’s something like: slowness is a signal of unseriousness, and being associated with it reflects on me. Or: if I’m not visibly the fastest thinker in the room, I’m not sure what I’m bringing.
That’s the generator. And notice that it isn’t a belief you’d defend — it’s not even a belief exactly. It’s closer to an appetite, a thing you’re hungry for, and the interrupting is one of several behaviours it produces.
Which explains something the first two levels can’t: why suppressing the behaviour makes it come out somewhere else. If you successfully stop interrupting but the generator is intact, you’ll find yourself doing something functionally identical — visibly checking the time, finishing emails with a summary of what the other person meant, becoming subtly cold with slower colleagues.
Same appetite, new outlet. People take this as evidence they have multiple faults. They have one, wearing different clothes.
Why it holds: when the generator changes, the impulse isn’t produced, so there’s nothing to suppress. You don’t stop interrupting through effort. It stops occurring to you.
Why This Order Matters Practically
The purist reading is: skip to level three, only level three works.
I think that’s wrong, and it’s the mistake that makes this framework useless to people who try to apply it.
Here’s the practical sequence:
Use one and two as containment while three is being built.
Level three is slow. It’s the accumulation of evidence about who you are, and it operates over months. During those months the behaviour is still being produced, and it’s still costing you.
So run vigilance where you can — meetings that matter, situations you’ve identified — knowing it’s a holding action rather than a solution. Run the reasoning work too, because it stops you defending the thing, and you can’t change something you’re still arguing for.
Just don’t mistake either for the work.
The failure mode isn’t using level one. It’s using level one and believing it’s level three — concluding, when it collapses under pressure, that you’ve failed at change, when actually you succeeded at containment and never attempted change.
That misreading is what produces the person who’s “tried everything” and has a decade of evidence that they can’t do this.
Finding the Generator
The obvious question: how do you find level three, given that the whole point is that it’s not what you’d say if asked?
Three moves, none of them clever.
Look at what the behaviour gets you. Not what it costs — that’s obvious, and it’s why you’re trying to stop. What does it deliver, reliably, in the moment? There’s always something. If there weren’t, the behaviour wouldn’t persist.
Look at the other outlets. What else do you do that produces a similar payoff? Three behaviours with one payoff between them means the payoff is the generator, and you’ve just found it by triangulation.
Look at when it’s absent. There are situations where the impulse doesn’t arise at all. What’s different about them? Frequently the answer names the appetite directly — I don’t interrupt people I’m not competing with is a considerably more informative sentence than anything you’d have produced by introspection.
Then the actual work is the identity mechanism: accumulating evidence, over time, that you’re someone the generator doesn’t describe. That’s stealing your future self, and it’s slow in exactly the way this framework predicts.
What This Is Not
It is not a claim that suppression is worthless. It’s useful, immediately, and sometimes it’s all you’ve got. A person about to go into a difficult meeting should absolutely use vigilance. The argument is against mistaking it for a permanent solution and then drawing conclusions about your character when it fails.
It is not an instruction to excavate your childhood. Finding the generator is a present-tense investigation — what the behaviour delivers now, where else it shows up now, when it’s absent now. Origin is interesting and mostly not necessary.
It is not fast. Level three is measured in months and produces nothing visible for most of them. Anyone offering a compressed version has removed the mechanism and kept the vocabulary, which is the standard trade in this genre.
The Cost
Two prices, and the second is the real one.
First: source-level work has no visible output for a long time.
Level one produces immediate results. You stop the behaviour today, you can observe it, and it feels like progress. That feedback is genuinely motivating and it’s why people keep returning to the level that doesn’t hold.
Level three produces nothing you can point at for months. You’re accumulating evidence, the accumulation is invisible, and there’s no week where you can tell it’s working. Then at some point the impulse simply isn’t there and you can’t say when it stopped.
Second: finding the generator is unflattering.
Level one lets you keep a decent opinion of yourself — you have a bad habit, you’re working on it. Level three requires naming an appetite, and the appetites underneath ordinary faults are rarely noble. Needing to be the fastest. Needing to be needed. Needing not to be dismissed.
Nobody enjoys writing that sentence about themselves, and the writing is the step. That’s the same cost as the audit in why positive thinking is the worst advice, and it’s why most people stay at level one for twenty years.
Common Questions
Why does a behaviour I’ve stopped keep coming back?
Because stopping the behaviour doesn’t remove what produces it. Suppression is a filter downstream of the impulse, and it’s staffed by attention, so it fails whenever attention or resource is disrupted. The impulse was never touched.
What’s the difference between suppressing a behaviour and changing it?
Suppression requires ongoing effort and ends when the effort is interrupted. Change means the impulse is no longer produced, so there’s nothing to suppress. The test is whether it takes effort — if it does, you’re suppressing.
Why doesn’t understanding why it’s wrong fix it?
Because the behaviour isn’t produced by the reasoning. The justification was constructed after the fact to explain something already happening, so defeating the justification leaves the mechanism intact. It also relies on reasoning being available, which it isn’t at the moments the impulse is strongest.
How do I find what’s actually generating it?
Ask what the behaviour reliably delivers in the moment, look for other behaviours that deliver the same thing, and examine situations where the impulse doesn’t arise at all. That third one usually names the appetite more accurately than introspection does.
Should I stop using willpower entirely then?
No — use it as containment while the slower work happens. The error isn’t using suppression, it’s believing suppression is the change and then concluding you’ve failed when it collapses under pressure.
Why does fixing one fault make another appear?
Because they share a generator. If you suppress one outlet while the underlying appetite is intact, it finds another. Several apparently separate faults with one payoff between them is usually one fault wearing different clothes.
Where to Start
One fault. Twenty minutes. Three questions, in writing.
Pick the fault you’ve most often tried to fix — the one with the longest record of failed attempts. That record is useful, not embarrassing: it’s evidence you’ve been working at the wrong level, which is a much better diagnosis than the one you’ve been carrying.
Then answer:
What does this behaviour reliably get me, in the moment? Not what it costs. What it delivers, every time, without fail.
What else do I do that gets me the same thing? List two or three. If you can, you’ve triangulated the generator.
When does the impulse not arise at all? Name a specific situation. Then say what’s different about it.
Don’t try to fix anything this week. The three answers are the work, because until you can name the generator you’ll keep attacking the output, and attacking the output is what you’ve been doing for however long this has been going on.
I’ve spent twenty-five years working on how ordinary people develop capabilities that look impossible from the outside, and the people who genuinely change a longstanding fault almost never describe having finally found the discipline for it.
They describe, usually with some surprise, that at some point it stopped occurring to them.