The Junk Drawer

Every home and office has one. You know which drawer I mean — the one with the takeout menus and the dead batteries and the rubber band that used to hold something together and the key that opens something, somewhere, you just can’t remember what. It is not a disorganized drawer. It is a perfectly organized drawer for exactly the things that don’t fit anywhere else — and for the few things that can fit in any other drawer, too. Those are not the same category, and the difference between them is what this series is about.

I’ve been thinking about junk drawers.

Not the kitchen kind, or not only that. What I mean is the structural role the junk drawer plays in any classification system: the place where you put things that resist classification, where the system acknowledges, implicitly, that it is not quite equal to the world it is trying to describe. But that is only part of the story. The junk drawer also serves as insurance against your own imperfect recall — you put something there not because it lacks a correct home but because you don’t trust yourself to remember where that home is, or you don’t trust whoever else uses the kitchen to look in the right place.¹ Game theorists have a name for this problem — imperfect recall — and the junk drawer is, in a precise sense, a structural response to it: a location whose value derives partly from the fact that everyone agrees to check it when they can’t remember where else to look. And then there is the third function, which is the most dangerous: the junk drawer as a honey-do list. You put something there because it needs attention, and the drawer is where things go when attention is deferred. This works, up to a point. Past that point, the drawer is too full to deal with, the deferral has compounded, and what you need is not a better classification system but a free Saturday and no other plans. I have a theory of spring cleaning and why nobody likes to do it. That is a different post.

Every filing system has a junk drawer. Every library has one. Every taxonomy, every ontology, every legal code, every agency budget has one — even the ones that don’t admit it. The question is not whether you have a junk drawer. The question is whether you know you have one, and what you do about it.

---

Consider the key.

Somewhere in your home — or in the home where you grew up, in a bowl by the door, or in the drawer itself — there is a key whose function is unknown. It is not unknown because it is unimportant. It is in the drawer precisely because it might be important. You kept it because the cost of being wrong about discarding it is higher than the cost of not knowing what it opens. The key is not unclassified because you failed to classify it. It is unclassified because classifying it correctly would require information that you don’t have yet — or that you once had and lost along with whatever it is the key unlocks.

Notice that the risk of guessing is not symmetric. If you place the key in the most plausible-looking drawer — the spare-keys drawer, say, or the one where the garage remote lives — and you are wrong, the key is now harder to find than it was in the junk drawer. The junk drawer is findable precisely because it is the designated location for acknowledged uncertainty: when you don’t know where else to look, you check there. A key filed under a confident but incorrect guess can only be recovered by accident. The junk drawer is not a failure to classify. For the lost key, it is the correct classification.

This is the first kind of junk drawer object: something important enough to keep, unclassifiable because the classification system hasn’t caught up with it yet. Or because it has, and you weren’t there to file it at the time. The system isn’t broken. The object isn’t defective. The gap is real and it sits in a drawer, and at least you know where it is.

Now consider the screwdriver.

You have a screwdriver in your junk drawer. So does your neighbor, despite organizing their kitchen completely differently. So does your friend who lives in a one-bedroom apartment and owns exactly one other tool (a hammer, for the pictures). The screwdriver is not in the junk drawer by accident or by laziness. It is there because the problems a household encounters — loose drawer handles, battery compartments, the occasional furniture assembly emergency at 11 p.m. — are roughly the same problems every household encounters, and the screwdriver is the right tool for most of them. A screwdriver dedicated to a specific, named drawer stops being the screwdriver-for-everything and becomes the screwdriver-for-those-particular-screws. Its value is its breadth. Classifying it precisely would diminish it.

This is what separates the screwdriver from the lost key. The key’s location in the junk drawer is temporary, contingent, waiting on information that might eventually arrive. The screwdriver’s location in the junk drawer is optimal. It is there because the structure of the problems you encounter makes it more useful unclassified than classified. You could, in principle, put a screwdriver in every drawer in the kitchen. The cost is space; the gain is marginal, because the junk drawer is right there. The location of the junk drawer turns out to be a design variable — a meta-organizational choice that shapes everything else. An optimally placed junk drawer renders redundant copies unnecessary.

If you have ever studied networks, you know what the junk drawer is. It is a hub: the node with unusually high connectivity, the location most frequently visited across the widest range of tasks. The screwdriver belongs there not because it is a hub but because it is optimally located at one. Its value derives from proximity to the place you already go when you don’t quite know what you need. The junk drawer functions as a hub in the classification network of your home. That is not a failure of the network. In a well-designed network, hubs are load-bearing.

There is a more general point here that will take more than one post to develop fully. The structure of the problems you face determines the optimal structure of your classification system — not just the labels on the drawers, but how many drawers you need, what goes in each one, and crucially, what goes in the junk drawer. The junk drawer’s contents are a kind of shadow of your problem space. Read them carefully and you will learn something about the problems your system was designed to solve and the problems it wasn’t.

---

Ordinary least squares regression — OLS, the most commonly used statistical technique in the social sciences — is a screwdriver. Ask a statistician where it belongs and they will tell you it is foundational to linear regression, which belongs in the statistics drawer. Ask an econometrician and they will tell you it is the workhorse of causal inference, which lives in the econometrics drawer. Ask a machine learning researcher and they will tell you linear regression is a special case of empirical risk minimization, which is a machine learning concept. A psychometrician will point out that factor analysis generalizes it. A political scientist will teach it in a methods course alongside logit and difference-in-differences, in a drawer labeled something like “quantitative methods” that itself isn’t quite a discipline — and alongside it, a drawer labeled “formal theory,” which is what happens when the quantitative methods drawer runs out of room and the political science department starts arguing about whether they’re the same drawer.

OLS is not in every discipline’s junk drawer because nobody bothered to classify it properly. It is there because its cross-categorical utility is the thing that makes it OLS. The statistician’s OLS and the machine learning researcher’s OLS are the same mathematical object applied to overlapping but distinct problem spaces, and the fact that each community calls it something slightly different — linear regression here, least-squares there, a special case of empirical risk minimization somewhere else — is not a failure of coordination. It is the junk drawer doing its job. Each discipline keeps a copy where it can reach it, and the copies are the same screwdriver.

---

Maggie and I have been thinking about and working on this kind of structure as a natural successor to our current work on classification and social outcomes — some of which has been showing up here over the past few weeks, in posts on what AI is and what accuracy-maximizing classifiers actually do. I will be talking about this project more over the next few weeks and possibly beyond.

What the key and the screwdriver have in common is this: both are fully accommodated by the junk drawer. That is the point. The junk drawer is not where classification breaks down. It is where a well-designed classification system acknowledges that some objects require a different kind of accommodation than a labeled compartment provides — one that is contingent (the key, waiting on information) or hub-like (the screwdriver, earning its ambiguity). A classification system that contains no junk drawer isn’t a complete system. It is a system that has mistaken the absence of acknowledged exceptions for the absence of exceptions, which is a different and more serious error. The junk drawer is not the symptom of a bad taxonomy. It is evidence that the taxonomy is honest about the world it is trying to describe.

Last week’s posts — on prediction markets and what it means to call something a swap or a bet or a securities contract — were, among other things, about the costs of refusing to maintain a junk drawer. The question of whether Kalshi’s contracts are futures or swaps or something else is genuinely hard, and the jurisdictional fighting over who gets to decide is in part a fight over who is allowed to build the drawer and what goes in it. A legal system that insists every instrument must be unambiguously classifiable before it can be regulated will find, reliably, that the instruments that matter most are the ones that most resist that requirement. The junk drawer fills up whether you admit it or not. The question is who controls it.

That is what this series is going to be about. Not just the junk drawer as a metaphor, but the junk drawer as a structural feature of classification systems — what kinds of objects end up there, why, what the difference is between a system that acknowledges its junk drawer and one that pretends not to have one, and what it means for democratic accountability and institutional design that the junk drawer tends to accumulate in exactly the places where the stakes are highest.

There is, as you might expect, a theorem lurking somewhere in here. There always is. The informal version goes like this: if you don’t have a junk drawer, eventually you’ll only have junk drawers. (Ed: is that a promise or a threat?)

With that, I leave you with this.

---

¹ There is a darker version of this dynamic that is worth naming. If the members of your household are evaluated — formally or informally — on how rarely they resort to the junk drawer, then “I don’t know where this goes” becomes a performance failure rather than an honest signal. The optimal response is to avoid the junk drawer regardless of actual uncertainty and place objects in the most plausible-looking available drawer instead. Readers who encountered the negative threshold rule a few weeks ago will recognize the structure: the incentive to classify confidently is strongest precisely when uncertainty is highest, which means the classification system receives its least accurate information exactly when accurate information is most needed. The neat freak’s kitchen is, on the surface, immaculate. Finding anything in it is a different matter.