Monday’s post introduced the junk drawer as a structural feature of any well-designed classification system — not a failure to be corrected, but a load-bearing component. It offered a typology: the lost key, which is in the drawer because it is waiting on information that hasn’t arrived yet, and the screwdriver, which is in the drawer because its cross-category usefulness is precisely what makes it valuable. And it closed with an aphorism in search of a proof: if you don’t have a junk drawer, eventually you’ll only have junk drawers.
Some readers may find that aphorism backwards. Surely the problem is too many junk drawers, not too few? Surely the goal of a well-designed classification system is to eliminate ambiguity, not to designate a drawer for it? This post is for those readers — and also for the ones who accepted the claim on Monday but want to see the mechanism. The Supreme Court helped illustrate it, in a ruling handed down last week that most people have been reading as a First Amendment story. It is also a story about the junk drawer problem at the foundation of law and governance — systems that operate, at every level, by classifying things.
Let me restate the formal claim in slightly different terms, because what follows will need it. A classification system that contains no junk drawer is making an implicit promise: every object that arrives at the system has exactly one correct home, and the system knows which one. This promise is almost never kept, because the world does not cooperate. Objects arrive that are genuinely cross-categorical — that have the properties of two or more drawers, or that have properties the system’s designers hadn’t anticipated. What happens next depends on whether the system admits the problem.
A system without a junk drawer faces a prior choice: acknowledge that some objects are genuinely cross-categorical and designate a location for them, or deny it and carry on. Systems that acknowledge it have the junk drawer problem — which is real, and what this series is mostly about. Systems that deny it face a different set of consequences, and for simplicity there are three basic ways one can manage day-to-day business while keeping the junk drawer hidden. The first is to refuse to classify ambiguous objects at all — but in a regulatory or legal taxonomy, “no classification” means “no jurisdiction,” and jurisdictions tend not to voluntarily disclaim themselves, so this option is almost never chosen. The second is to assign the object quietly to the nearest available category without acknowledgment — a locally defensible placement that accumulates silently over time, invisible in any single instance, transformative in aggregate. The third is to expand the definition of an existing category explicitly to absorb the newcomer — at which point that category is doing the work of a junk drawer while formally denying it.
In legal systems, the third option tends to dominate, because legal systems require explicit justification for their classifications. The justification takes the form of expanding a category’s stated criteria. The category becomes a contested site rather than a settled one. At that point you don’t have a taxonomy with clean drawers and one acknowledged junk drawer. You have a taxonomy where every drawer is, functionally, a junk drawer: a site of active classification dispute rather than settled assignment. The ambiguity has not been eliminated. It has been distributed.
For roughly eighty years, American courts evaluated economic regulations — including occupational licensing requirements — under a constitutional standard so deferential it was nearly invisible. The governing rule came from a 1955 case called Williamson v. Lee Optical, which held that a legislature’s economic regulation need only be conceivably rational to survive constitutional challenge. It does not have to be actually rational. It does not have to be effective. It does not have to serve the purpose the legislature claimed. The legislature might have enacted a needless, wasteful requirement in many cases, and the requirement could still be constitutional. The courts’ job was to imagine whether a reasonable legislature could have thought the regulation served some purpose — and then to find it constitutional on that basis.
What this standard created, in practice, was a junk drawer for professional regulation. Licensing boards — controlled, as often as not, by the people who benefit from reduced competition — obtained statutes that made it illegal to give certain kinds of advice for money without a government license. Courts upheld these statutes routinely, on the theory that giving advice is “conduct,” and states may regulate conduct. The “conduct” drawer, under rational basis review, had no bottom. Hair braiders. Interior designers. Florists. Nutritionists. The screwdriver and the lost key and the dead batteries all went in together, and courts said: the legislature could have thought this was a good idea, so it passes.
Last week, in Chiles v. Salazar, the Supreme Court said something that had been building in the lower courts for years.1 Writing for eight justices, Justice Gorsuch held that when a state restricts purely verbal, paid advice — telling a client what to eat, how to manage their grief, what spiritual path to consider — it is not regulating conduct. It is regulating speech. And calling advice “conduct” to escape First Amendment scrutiny is, in his words, “just a labeling game.” The courts had been accepting that label for eighty years. The Court has now noticed that it is a label.
The immediate consequence is that a large number of professional licensing laws — those that restrict information rather than action, advice rather than procedure — are now vulnerable to First Amendment challenge. The 11th Circuit had upheld Florida’s dietetics licensing law in 2020 using exactly the “it’s conduct, not speech” framework. That analysis, the Court made clear, no longer holds. The professions with the most exposure are those whose licensing laws restrict verbal advice given to willing adult clients with no strong historical tradition of state regulation. There are many of them, and what they share is that the regulatory authority over them was parked in the “conduct” drawer not because anyone thought hard about whether that’s where it belonged, but because the drawer had no bottom and accepted anything labeled correctly.
Notice what has happened at the system level. The “conduct” drawer was doing the work of a junk drawer — holding cross-categorical objects that the taxonomy didn’t have a better home for — while explicitly denying that it was one. Professional speech is both conduct and speech. It has properties of both. Forcing it entirely into the conduct drawer, and then deferring to whatever the legislature said about it, was a way of refusing to acknowledge the ambiguity. The drawer filled up. Courts accepted the label for decades because they had also accepted that the assignment was the legislature’s to make.
What Chiles does is contest the label. It says the drawer assignment is not purely the legislature’s choice — that there is a prior question of what kind of thing professional advice actually is, and that question has a constitutional answer. The consequence is not that professional regulation disappears. The consequence is that the ambiguity migrates. Every licensing regime that restricts advice is now a potential First Amendment case. The First Amendment drawer — which was designed for press freedom and political speech and artistic expression — is now being asked to adjudicate whether a state can require a license to tell someone which vegetables to eat. The drawer is being expanded. It is doing work it was not designed to do.
The mechanism behind this is social before it is architectural. A junk drawer functions as a coordination device: it works because everyone who shares the space knows it’s there, which means everyone knows where to look when they don’t know where to look. A private junk drawer — maintained by one person, unknown to the others who share the space — solves nothing. The coordination problem requires the acknowledgment to be shared. The gradual colonization of clean drawers by cross-categorical objects happens precisely because there is no shared location for acknowledged uncertainty. Each reassignment is individually defensible: the adviser went in the conduct drawer because that is where regulated activities go; another practice went in there for the same reason; then another. No one placement looked like a mistake. Collectively, the assignments were transformative. The drawer filled with objects placed there not because they belonged but because there was nowhere else to put them, and no one was saying that aloud. The misclassifications accumulated silently, one plausible-seeming placement at a time, until the drawer’s label stopped describing its contents and the people opening it were working from different mental maps of the same office — each confident in where they left something, none of them sure what the others find when they look.
A label does not enforce its own accuracy. It describes what the drawer was designed to contain — at the moment of its creation, or the last time someone examined it carefully — and it goes on describing that, regardless of what accumulates afterward. When enough cross-categorical objects have settled inside, the label stops being a description and becomes a location marker: the name people use when they don’t know where else to look. At that point the drawer has been relabeled in practice, through use, even if the sign on the front has not changed. A drawer that everyone in the office refers to as the random-stuff drawer is a junk drawer. It is just not honestly labeled as one. The drawer labels its contents; the contents, eventually, label the drawer.
This is the pattern this blog has been tracing for several months now, in several different contexts. Executive tariff authority under IEEPA was struck down; the problem migrated to Section 122. The jurisdictional fight over Kalshi’s prediction market contracts was a fight over whether CFTC’s “futures” drawer or state-law “gambling” drawers had authority; the answer was constitutive, not descriptive — the label determined the regulatory reality. In each case, the observation is the same: the impossibility of clean classification is not resolved by refusing to acknowledge it. It is conserved, and it migrates to wherever the pressure finds least resistance.
None of this is new. Regulatory agencies are themselves junk drawer objects in the American constitutional taxonomy, and they have been since the taxonomy first encountered them. The Constitution distributes governmental power among three mutually exclusive drawers — legislative, executive, judicial — and takes some care to specify that they do not overlap. Agencies exercise authority from all three: they write rules with general applicability, enforce those rules against specific parties, and adjudicate disputes about that enforcement. The taxonomy has no drawer that fits them. American constitutional law has spent roughly 140 years not quite admitting this.
The institutional origin of the problem is precisely traceable. By the 1870s, more than a dozen states had built railroad commissions to address freight rate discrimination — a new kind of institution with no clean constitutional name, installed wherever the existing drawers didn’t reach. Then in Wabash, St. Louis & Pacific Railway Co. v. Illinois (1886), the Supreme Court held that states could not regulate rates for the interstate portions of rail journeys. This did not resolve the problem of railroad rate discrimination. It migrated it. Congress responded the following year with the Interstate Commerce Commission — copying the state commission model directly, ambiguity and all — and moved the junk drawer one level up in the federal system. When the Supreme Court eventually needed a word for what agencies actually are, it produced “quasi.” The Federal Trade Commission, it held in Humphrey’s Executor v. United States (1935), is “quasi-legislative and quasi-judicial.” Quasi means resembling but not actually. It is the label you write on a drawer when you are not prepared to admit you have one.2
The non-delegation doctrine is the recurring constitutional attempt to close this drawer — to require that Congress, when creating agencies, supply an “intelligible principle” constraining agency discretion, thereby re-anchoring agency authority in the legislative drawer where it nominally belongs. The debate has been live for nearly a century and shows no sign of settling. The formal reason is not difficult to see from where we are standing: any intelligible-principle standard specific enough to actually constrain agencies also forbids delegation in exactly the cases where delegation is most necessary — when the regulatory problem is too technically complex or too fast-moving for Congress to resolve in advance. Any standard permissive enough to allow necessary delegation permits essentially everything. The criterion for “how much delegation is too much” is itself a junk drawer object. The non-delegation debate does not recur because courts keep applying the wrong test. It recurs because the right test cannot be specified without acknowledging the drawer. My work with Sean Gailmard — including our book Learning While Governing — has argued that this difficulty is structural rather than merely doctrinal: agencies develop expertise and policy knowledge over time in ways that cannot be reduced to what Congress specified at the moment of delegation, and the legitimacy question that creates is precisely the question the taxonomy was built to avoid. There is a full argument to make here. It belongs to this series, and we will get to it.
A classification system with an honest junk drawer does not have this problem — or not in the same form. It has a different problem, which is the problem of what goes in the drawer, and what the rules governing the drawer are, and who controls it. That problem is real and worth taking seriously, and it is what this series is going to spend most of its time on. But it is a more tractable problem than the one you get when you pretend the drawer doesn’t exist. The question of how to manage acknowledged uncertainty is hard. The question of how to manage distributed uncertainty, once it has contaminated every drawer in the taxonomy, is harder — and the answer is usually something like a free Saturday and no other plans.
So here, stated more completely, is the organizing claim the series is building toward. A classification system that forbids acknowledged ambiguity does not thereby eliminate ambiguous objects. It forces them into existing categories. Categories that absorb objects outside their design criteria become internally contested. Internally contested categories are, functionally, junk drawers: they are the locations in the taxonomy where you go when you don’t know the right answer and need somewhere to look. A classification system whose designers refuse to designate a junk drawer will, as it encounters the world, fill every drawer with the objects that don’t fit — until the taxonomy as a whole is a set of contested, overcrowded, and only loosely constrained categories. That is what junk drawers look like when they are evenly distributed rather than honestly acknowledged.
If you don’t have a junk drawer, eventually you’ll only have junk drawers. The theorem requires the full version: what matters is not merely whether the drawer exists, but whether its existence is shared knowledge. A classification system that maintains a private location for acknowledged uncertainty — known to its designers, invisible to its users — has not solved the coordination problem. The drawer only does its work when everyone who uses the system knows to check it. Which means the relevant failure mode is not just the absence of the drawer. It is the absence of the announcement.
The next post in this series will start asking about the inside of the drawer: given that we have one — that every well-designed system has one — what are the different kinds of objects that belong there, and what follows from the difference? The lost key and the screwdriver are not the same kind of problem. The distinction will matter.
With that, I leave you with this.
1 I am drawing on the general principle from the ruling and from the Hill’s summary of its scope; verify case name and citation before publishing. The key holding — that calling advice “conduct” to escape First Amendment scrutiny is a labeling game courts need not accept — is the operative point regardless of the specific facts. The Gorsuch opinion was joined by eight justices, including Justice Jackson, which means this is not a partisan ruling in the conventional sense. That is worth noting somewhere in the text if there is a natural place for it.
2 The “quasi” language in Humphrey’s Executor has never been fully explained or abandoned. It survived Morrison v. Olson (1988), was narrowed by Seila Law v. CFPB (2020) and Collins v. Yellen (2021), and remains in contested use. Each round of litigation relocates rather than resolves the ambiguity — which is, at this point, a fairly strong empirical confirmation of the theorem.