I have stood in a doorway outside a seminar room and heard a colleague say he’d thrown the slides together that morning. I have heard some version of it many times. I assume I have said it myself, and here is the part I want to be honest about: I can’t produce a single instance. Not one talk, not one doorway. The sentence is usually true, which is what makes it work, and it is automatic enough that it leaves nothing behind in the person saying it.
It leaves something behind in the room. A graduate student once told me, and I’m compositing several conversations here, that it was reassuring to hear someone senior still worked that way.
So the sentence does two things, and neither of them is a decision. It makes sure that if the talk goes badly, an explanation is already standing by that isn’t about the speaker. And it tells a room with no way to check how mathematics gets made, and the room believes it.
That asymmetry is most of what I want to say here. The sentence costs the one person who could stop saying it nothing at all.
The bet that pays either way
This has a reasonably clean origin. In 1978, Edward Jones and Steven Berglas gave people a test. Half got solvable problems and were told they’d done well; success they could account for. The other half got problems that were secretly insoluble and were also told they’d done well: success they had no way to explain and no reason to think they could repeat. Everyone was then offered a choice of drug before a second test, one described as performance-enhancing and one as performance-inhibiting.
The people handed a success they couldn’t explain disproportionately chose the drug that would make them worse.
The mechanism is attributional. Put an obstacle in front of yourself and look at the possible explanations. Fail, and low ability gets discounted, the obstacle accounts for it. Succeed, and high ability gets augmented, you did it anyway. You have reduced the variance on the estimate of your own ability and paid for it in output. It’s insurance, and the premium is real work.
The precondition is what most retellings get wrong. This isn’t driven by thinking you’re bad. It’s driven by holding a picture of yourself you can’t account for and don’t trust yourself to reproduce: something worth protecting, plus genuine doubt that performing will protect it. That sits on the attribution side of a line I’ve drawn here before, not the calibration side. Not how good am I, but what will this be taken to mean about me.
A word on the evidence, and then I’ll leave it alone, because unlike Dunning-Kruger this isn’t a case where the state of the literature should change what you do. The paradigm is old and small, and field measurement runs mostly on self-report; a weak instrument for a behavior whose function depends on you not seeing it clearly. None of that matters much, because you can locate an instance in your own last six months. If you can’t, that’s worth sitting with on its own.
Two forms, and only one is expensive
The doorway version is a claimed handicap: you assert the impediment without creating it. I only looked at this last night. I’ve been buried this semester. This is very preliminary. Cheap, near-universal, and almost as protective as the real thing at a fraction of the price.
The behavioral version builds the obstacle for real. You don’t rehearse. You take on a fourth commitment during a year when the mathematics isn’t going anywhere. You start the write-up the night before. Here the handicap actually degrades the work, which is the point. One that costs nothing isn’t credible.
They’re less separate than they look, and the bridge is the thing I’d most want a younger version of myself to know. A claimed handicap has to be maintained to stay credible, and maintenance is behavioral. If your standing account of yourself is that you’re slow at writing, or bad at talks, or hopeless with anything computational, the cheapest way to keep that account true is to keep not doing those things. Given enough years the excuse becomes a fact about you, with documentation.
Three things this is not
Defensive pessimism produces the identical sentence in the hallway and the opposite behavior afterward: announce disaster, then prepare harder than anyone. The anxiety is being converted into preparation, and for those people it works. You can tell the two apart only by what happens in the next four hours.
Procrastination is usually task aversion, and becomes self-handicapping only when the delay is doing attributional work.
Impostor syndrome is the felt state; this is one strategy that state produces.
Why Mathematics is good ground for it
In most professions, succeeding despite an obstacle earns you a modest bonus. In mathematics it’s a status claim on its own, because the field carries a live mythology of effortless gift; the prodigy, the person who either has it or doesn’t. I’ve written about how a fixed-mindset belief inverts the meaning of effort, so that trying reads as evidence against you rather than for you. That inversion is exactly what makes not-appearing-to-try pay twice here. The handicap doesn’t only protect you from a bad inference. It buys a good one.
And the shape changes as you go.
For a student: the qualifying exam you let everyone know you didn’t study for, which is where most of us first learn the move. The chapter you don’t show your advisor, because an unread draft can still be good.
For a postdoc: the job talk given cold, on purpose. The preprint held off the arXiv until the applications are out, so it can’t be counted against you and can still be called forthcoming.
For faculty: the service commitment accepted in the middle of a stalled year, which feels like generosity and functions as an alibi. The finished result that never gets written up. The new subfield entered just far enough to have a reason for not producing in it yet. The collaboration where your particular contribution stays comfortably hard to isolate.
And then the version that arrives with security rather than in spite of it. After tenure the obstacles get larger, more admirable, and much harder to question. You chair the committee. You run the seminar, the colloquium, the departmental review. All of it is real work, genuinely needed, and it is also the most unassailable alibi in the profession; nobody has ever been asked why their paper is two years into revision while they’re chairing the search. There is no one left whose job it is to notice, which means the only check is the one you run yourself, at the stage when you’ve had the least practice.
What it does downstream
It breaks the instrument. You hedged because you couldn’t tell whether you could repeat a success, and the hedge guarantees the next outcome won’t tell you either. I’ve argued before that failure is the high-information event in a research career: it rules something out cleanly, while a success stays low-information unless you do the work of interpreting it. You can’t check an estimator against outcomes you’ve been corrupting on purpose. And the shorter your window, the more each muffled outcome costs. Someone with three years to assemble a record strangers will read can’t spare many.
That last point cuts both ways, and it’s worth conceding. On the job market the hedge is partly correct: you’re one of eleven files read in an hour, and sounding underprepared has a cost private calibration won’t offset. But the presentation you give the room and the reading you take for yourself are two different instruments, and letting the first destroy the second is the avoidable half. Say this is preliminary out loud if you need to. Write down privately what you actually expected.
It manufactures the norm that makes it pay. Nobody has ever said I rehearsed this eleven times in a hallway, so the public record of how mathematics gets produced is assembled entirely out of disclaimers. The junior person in the doorway isn’t hearing a hedge; they’re hearing a data point about the production function, from someone with authority, with no way to check it. They conclude that good work comes from people who don’t prepare, that visible effort is something to conceal, and, the loop closing, that the claimed handicap is right for them too.
And it taxes the honest. A pre-emptive excuse and a well-calibrated statement of risk arrive on the same channel and look nearly identical on it. There are real difficulties with this approach is sometimes the most accurate sentence in a proposal and sometimes an escape route laid in advance. A panel can’t invert that map, already the panel’s central problem, and the more hedging there is, the cheaper honest hedging is to imitate and the less it is worth. The handicapper isn’t only spending their own output.
And it runs downhill. The version worth naming is the one nobody doing it experiences as a move: an advisor who says I haven’t had a chance to look at your draft has handed a student an obstacle they didn’t choose, can’t name, and will very likely absorb as information about their own work rather than about the shape of someone else’s week.
How you would actually see it
The obstacle carries no signal. Two people miss the same deadline with the same true explanation and one of them is handicapping. So every usable test keys on something other than the obstacle, and the design is always the same: you can’t run a between-subjects comparison on yourself, so hold yourself fixed and vary the evaluative stakes.
Four of them, in rough order of how much real data they give you.
Where the stall happens, not whether it happens. Take one project, one is enough, and locate the step you keep not doing. If the difficulty were mathematical, you’d stall at the mathematics. If instead you stall at the transitions into evaluability,the proof that doesn’t get written up, the draft that doesn’t go to your advisor, the submission that doesn’t happen, the revision that sits for five months, the binding constraint isn’t difficulty. It’s exposure.
Where the week actually went. The other tests look for a decision. This one looks for the absence of one. Nobody decides not to prepare; you decide, a dozen times across a week, to do other things, answer the mail, take the meeting, referee the paper, fix the figure, and every one of those is defensible on its own, which is exactly why the result has no decision point you can locate afterward. So don’t ask what you chose. Ask where the hours went in a week with something evaluated at the end of it. You rearranged three things to get to a conference you wanted to attend. You found an afternoon for a colleague’s draft. Somehow there was no two-hour block for the talk. Same week, same calendar, same person; the resources were there, and they were allocated away from the one thing that was going to be judged.
Obstacle timing against the evaluation calendar. Real constraints don’t know when your talk is. A sick child, a visa appointment, a roof; these arrive on a schedule uncorrelated with when you’re being judged. A handicap has to be in place beforehand to do its work. One instance means nothing; twenty checked against a calendar you already keep is a genuine measurement. If you haven’t accumulated twenty, start the list now. It’s the only test here that costs nothing but time.
The pre-written explanation. Before the outcome exists, write down two things: your honest probability it goes well, and the sentence you’d say if it doesn’t. If that second sentence is already fluent, already fully formed before you know anything, the explanation predates the evidence. An explanation that arrives after an outcome is inference. One that arrives before it is architecture.
The people this does not apply to
If your obstacles track your life rather than the evaluation calendar, they’re real, and that is the finding. Someone teaching a 4/4 with a parent in hospice doesn’t need a psychological construct to explain a thin year. A framework suggesting your obstacles might be self-inflicted is a dangerous gift to anyone already inclined toward self-blame, and the people who’ll apply it hardest are exactly those carrying the most genuine, least chosen constraints. There’s no guilt available here and none should be taken. Knowing a constraint is external is worth having on its own, because it stops you spending effort on a lock effort doesn’t open.
And there’s a third case, neither handicap nor imposition, that gets misfiled constantly. Suppose you limit travel to be home with your family. Ten years on, your collaboration network is thinner than it would have been. That isn’t self-sabotage. It’s a fair price paid knowingly for something chosen on the merits, and if you’re making that choice now rather than looking back at it, the useful thing to know is that the cost is real and the choice can still be right. What you want to avoid is paying the price and then later reclassifying it as a failure of nerve.
It converts into a handicap only in the telling. I don’t travel much, offered as a priority, is one thing. I would have been at that workshop, but I don’t travel, offered pre-emptively to explain an absence from a collaboration, is doing different work with the same true fact. And the hard case, which I’d rather name than skip: if you’d have declined that trip anyway, because you’d have had to speak and weren’t sure about the result, the family reason is still true, and it’s now also the acceptable public face of an exposure decision. Self-handicapping has never required the obstacle to be fake. Truth is what makes it work.
The test still applies. Did you decline uniformly, or did you make it to the meeting where you weren’t speaking and miss the one where you were?
Taking the obstacle off
Detection and removal are different problems, and removal splits by form.
For the claimed version, the fix is a speech act: stop issuing the disclaimer. It costs you the escape route, which is the whole price and isn’t nothing. The substitute does more work anyway; I’m not confident about section 3 invites help, while I threw this together forecloses it.
For the behavioral version, the obstacle is on a calendar and had a decision point you can usually locate. Decline it, renegotiate it, or finish it and schedule the preparation before you know how much is riding on the outcome, so it isn’t available later as a variable to adjust. Junior people are often offered choices that aren’t real ones, and that constraint is genuine. But people who could have declined routinely remember themselves as having had no option, and the memory is worth checking against the calendar.
Stop maintaining the identity claim. If you’ve kept I’m bad at talks true by not giving them, the removal is one talk you actually prepare. Not a resolution about the future. One instance.
Get the record from someone who has it. You can’t see the recurrence yourself, because each instance felt situational and they don’t accumulate anywhere you can look. Someone who’s watched you for five years can just tell you whether you say the same sentence before every talk. If nobody has watched you that long, an officemate or someone you’ve shared a seminar with for two years will do. The requirement isn’t seniority. It’s a longer memory of you than your own.
And if you have any standing, spend it. Say I rehearsed this six times. Return the draft on time, or say plainly why you didn’t. The disclaimer is cheap for you and expensive for the room, and you’re positioned to pay that difference. Standing here is relative: a fourth-year student has it over a first-year, and the first-year is listening at least as carefully as anyone listens to me.
One clean measurement
I don’t think the conclusion is to strip out every hedge, and I’d distrust an essay that said so. Some of them are load-bearing. I tried as hard as I could and it wasn’t enough is a sentence with no exit, and someone who’s never left themselves a way out may just be someone who hasn’t needed one yet.
So the narrow version. Take one thing and run it clean, no announced obstacle, no pre-written explanation, prepared the way you’d prepare if nobody were going to ask how long it took. Not the biggest thing. If everything on your calendar is consequential, take the smallest consequential thing, because the reading you’re after doesn’t depend on the stakes being high. A twenty-minute seminar talk returns the same measurement as a job talk and costs far less to get wrong.
Then look at what it gave you, decide what to do differently, and do it again next year. That’s the method. It isn’t a cure, because the conditions that produce the hedge, a self-image you can’t fully account for, an outcome you can’t control, a room whose judgment you can’t observe, are the standing conditions of a research career. They don’t go anywhere. You re-check the estimator because the estimator drifts.
I still say things in doorways. I’ve started hearing myself do it, which is new and isn’t the same as stopping.
And I still can’t give you the instance. That’s the part I’ve come to think matters most. No single afternoon of this ever cost me anything I could point to, which is exactly why there’s no memory of it to retrieve, the whole transaction is designed to leave the speaker unmarked. But somewhere there is a graduate student who heard it and filed it, quite reasonably, as evidence about what this work looks like from the inside. In ten years they’ll be in a doorway of their own, saying something like it to someone younger, and it will be true then too, and they won’t remember saying it either.