On July 14, 2025, in McMahon v. New York, the Supreme Court issued an order staying an injunction that would have prevented the Secretary of Education from implementing a Reduction in Force that would have cut the Department of Education’s work force approximately in half. Translated into English: the Secretary of Education announced layoffs, lower courts tried to stop the layoffs, and the Supreme Court stopped the lower courts from stopping the layoffs. The Supreme Court gave no reason for its order and did not respond to Justice Sotomayor’s 19-page dissent, which Justices Kagan and Jackson joined.
Why did the Supreme Court stop the stop, and was its decision correct? It is tempting to attempt a viral post raging that one side or the other was lawless, but the post won’t write. The best I can do is a milquetoast post observing that both sides had reasonable arguments. Sorry.
The interesting question for me is: why was this case, and so many other shadow-docket cases, decided by a 6-3 vote? Is it because law is just politics? Or because the government’s arguments were aligned with traditional conservative jurisprudential commitments, and the challengers’ arguments were aligned with traditional progressive jurisprudential commitments? Or a third explanation? That is the question I will explore in today’s post.
Airplanes, baseball, and the Department of Education
Little-known fact: Congress created the Department of Education back in 1867! A year later, however, the Department was demoted to a mere Office. For the next 111 years, Americans engaged in bitter debate over whether Education should be managed by a Department or an Office. In 1979, the Departmentalists finally prevailed, and Department of Education 2.0 was born.
Today, the Department of Education runs the federal student-loans program, administers K-12 grants, and enforces various education-related statutes such as Title VI (barring race discrimination) and Title IX (barring sex discrimination). You might remember the Department of Education from such Biden-era losses as Biden v. Nebraska, in which the Supreme Court rejected the Department of Education’s student-loan cancellation program, and Department of Education v. Louisiana, in which the Supreme Court declined to stay lower-court orders that threw out all of the Department of Education’s new Title IX rules.
Early in his Administration, President Trump described the Department of Education as a “big con job” and said he would “like to close it immediately.” When he nominated Linda McMahon as Secretary of Education, he asked her to “put herself out of a job.” On March 11, 2025, shortly after being confirmed, Secretary McMahon issued a memorandum that began with a bang: “As part of the Department of Education’s final mission, the Department today initiated a reduction in force (RIF) impacting nearly 50% of the Department’s workforce. Impacted Department staff will be placed on administrative leave beginning Friday, March 21st.” The memo noted that the Department’s workforce would be decreased from 4,133 to 2,183 workers. The memorandum promised, however, that the “Department of Education will continue to deliver on all statutory programs that fall under the agency’s purview.” Nine days later, President Trump issued an Executive Order declaring: “The Secretary of Education shall, to the maximum extent appropriate and permitted by law, take all necessary steps to facilitate the closure of the Department of Education.”
What do I think about this? On the one hand, the Department of Education seems to engage in many worthwhile initiatives. On the other hand, important American innovations such as the airplane, baseball, and the Department of Education itself were conceived without the benefit of the Department of Education. In conclusion, I have no opinion whatsoever about the optimal size of the Department of Education. I do have an opinion that the Secretary of Education cannot shut down the Department of Education when Congress has mandated that it continue to exist. But despite the memorandum’s grand opening that the Department was about to implement its “final mission,” the memorandum asserts that the Department will continue to accomplish all of its many mandatory statutory missions. And because everything the government says is true, that is that. Right?
In any case, as with all executive actions from now until the end of time, a maelstrom of litigation ensued. The plaintiffs were blue states, local governments, and teacher’s unions. Their argument was straightforward:
The Department of Education must, by statute, do various things.
The plaintiffs rely on the Department to do those things, and will be harmed if the Department doesn’t do them.
If the Reduction in Force is implemented, the Department won’t be able to do those things anymore:
You can’t cut 50% of the workforce and expect the Department to continue doing all the things it’s supposed to do.
The President and Secretary have announced that they’re going to shut down the Department, so it’s reasonable to infer that they’re going to shut down the Department.
Because the Reduction in Force will render the Department unable to do mandatory things, it is illegal.
Therefore, the court should enjoin the Reduction in Force to ensure this illegal thing doesn’t happen.
I’m simplifying a little, but only a little. A federal district court in Massachusetts agreed with this argument and enjoined the Reduction in Force. The First Circuit allowed the injunction to stand. The government took the case to the Supreme Court, which stayed the injunction and allowed the Reduction in Force to proceed.
Why?
We don’t know why the Supreme Court stayed the injunction, but there are a few reasons it might have done so.
The first is standing. The government said the plaintiffs lacked standing because their claims of injury rested on speculation: nobody really knew what was going to happen, and no particular plaintiff could prove that any particular bad thing was going to happen to it. The plaintiffs retorted that when you fire half of the Department of Education’s employees, of course bad things are going to happen.
Well … you can see both sides of this. From the government’s perspective, it just isn’t the case that More Employees = Better. Remember when Elon Musk fired 80% of Twitter and nothing happened? And how about Brooks’s Law (“Adding more manpower to a late software project makes it later”)? That is true of software, most definitely true of litigation teams, and perhaps also true of government agencies. If the Reduction in Force went into effect, some Department of Education operations might slow down, but neither the Constitution nor Magna Carta prescribe that “the optimal speed of the Department of Education is the speed associated with 4,133 as opposed to 2,183 workers.” And no plaintiff actually knew which operations would slow down, and “something bad will happen to someone” isn’t a basis for an injunction.
On the other hand, the plaintiffs made arguments such as: “The RIF has eliminated most of the department staff who handle certifications and recertifications, including seventeen out of the eighteen financial analysts and the entire team that oversees large school groups and foreign schools that participate in Title IV programs. Analysts with decades of experience attested that it will be impossible for the Department to perform its certification functions with the staff who remain, and that outside staff lacks the expertise to perform those functions.” Gulp.
This is a hard issue. Well, maybe not “hard,” exactly. It’s not hard in the sense that an International Mathematics Olympiad problem is hard. A better word is discretionary. The plaintiffs’ theory is somewhat speculative. Whether it’s too speculative to justify an injunction is a pure discretionary judgment call.
Let’s move on to issue #2: district court jurisdiction. When a federal employee is fired, he can’t sue directly in district court. Instead, he has to follow a hideously complicated administrative process that possibly involves the Merit Systems Protection Board and/or the Federal Labor Relations Authority.
Here, however, the plaintiffs weren’t the affected employees. Instead, the plaintiffs were third parties that said they would be harmed by the affected employees’ firings. And so, the plaintiffs argued, they didn’t have to follow that administrative process and could proceed directly to district court.
The government argued that the plaintiffs were being too clever by half. It would be weird, the government observed, if an employee who wanted to sue over his own firing had to navigate this gauntlet, but a third party who wanted to prevent that same employee from being fired could go to district court and get an insta-injunction. The government inferred from this weirdness that there’s no such thing as a lawsuit challenging the firing of an employee that isn’t you. According to the government, Congress provided one, and only one, vehicle for a court to undo a firing—a lawsuit by the employee himself who follows the administrative process that Congress prescribed.
The plaintiffs responded that the government was reading things into the statutes that aren’t there. There is no law that actually says a third-party plaintiff can’t challenge a Reduction in Force. The Reduction in Force is final agency action; the plaintiffs are harmed by that final agency action; therefore, under the Administrative Procedure Act, they may sue. Also, the plaintiffs weren’t challenging individual adverse employment actions. They were challenging the shutdown of the Department of Education’s functions, which the Department was effectuating via the mass firing of the employees who perform those functions.
Again, this issue could go either way. It does seem weird that a third party could persuade a court to enjoin a mass layoff, but whether that’s weird enough to warrant departing from the literal words of the relevant statutes is a discretionary call.
Issue #3 is the scope of the injunction. The district court enjoined the Reduction in Force in its entirety. The government argued that this was unnecessary and overbroad: the plaintiffs couldn’t possibly show that reinstating every single affected employee was necessary to prevent irreparable harm. The plaintiffs argued that the court wasn’t in the position of picking and choosing which employees would lose their jobs, so the only thing it could realistically do to give the plaintiffs relief was to enjoin the entire Reduction in Force. The government responded that litigation doesn’t work that way. It’s the plaintiffs’ job to propose a tailored injunction. If the plaintiffs can’t do that, they don’t get an injunction. Sure, the government wasn’t making it easy for the plaintiffs, but it’s not the government’s job to tie up executive action in a neat little bow to make it easier for federal judges to issue injunctions.
This dispute tees up a frequently-arising question: Go Broad or Go Narrow? Even assuming the Reduction in Force would result in statutory violations, there was no perfect remedy here: any remedy would either be too broad (reinstate every single employee who was laid off) or too narrow (issue some weak directive requiring the Department of Education to follow the law, which the Secretary claimed it would do anyway). There’s no objectively right answer to Broad Versus Narrow. This decision required discretion yet again.
The 1916 New York Giants
The Administration’s victory in McMahon v. New York is part of a broader pattern.
By my count, the Supreme Court has decided 18 shadow-docket applications filed by the Administration this term.1 The Administration’s first such application, Bessent v. Dellinger, resulted in a confusing 5-2-2 decision holding the case in abeyance for a few days, and the case then became moot. In the second case, Department of State v. AIDS Vaccine Advocacy Coalition, the Court denied the government’s application by a 5-4 vote. Since then, if one sets aside the Kilmar Abrego Garcia case, in which the Court issued an ambiguous decision that prompted both sides to declare victory, the Administration has prevailed every time—15 times in a row. We’ll have to wait and see whether the Administration can best the 1916 New York Giants’ 26-game winning streak.
Voting patterns have recurred. Justices Thomas, Alito, Gorsuch, and Kavanaugh have voted for the Administration all 15 times—those four Justices also dissented in AIDS Vaccine Advocacy Coalition, so call it 16 straight times—while Justice Jackson has voted against the Administration in all 16 of those cases.
Why has the Administration been so successful? One argument I’ve heard is that the Administration is bringing cases to the Supreme Court only when the lower court’s opinion is weak and the Administration is confident of victory. There is an element of truth to that, but it is not the full truth. To my eye, the Solicitor General has raised the cases that are most important to the Administration, not merely the easiest ones. Also, none of the Court’s orders resolving the Administration’s applications (apart from Abrego Garcia) have been unanimous, indicating that they haven’t been that easy.
So what is happening here?
Well, a cynic would say that All Judging Is Politics, and Republican appointees vote for Republican administrations while Democratic appointees vote for Democratic administrations. I don’t agree with that assessment.
An idealist would say that all 16 applications have teed up pristine legal issues squarely implicating cleavages between conservative and progressive legal philosophies, so every member of the Court in every case is merely applying pre-selected neutral principles and following the law where it leads. I am not sure that captures the dynamic, either.
Occam’s Razor
I am going to digress to a discussion of a case that has always bothered me: the student-loans case from two years ago, Biden v. Nebraska. I promise I am going somewhere with this.
The primary question presented in Biden v. Nebraska was whether the President was authorized by statute to cancel $430 billion in student loan principal. No, the Supreme Court held, by the ubiquitous 6-3 vote. As I see it, Biden v. Nebraska was a genuinely difficult case where the arguments for the challengers and the government lined up neatly with conservative and progressive jurisprudential principles respectively, so it’s no surprise that the case wound up 6-3.
Biden v. Nebraska also presented a threshold issue of standing. For opponents of the loan cancellations to bring a lawsuit, they needed to find a plaintiff. This wasn’t so easy. Lots of people were philosophically opposed to the student loan cancellations and wanted to sue, but philosophical opposition isn’t a basis to go to court. Because the case involved the government’s cancellation of billions of dollars worth of student loans, the primary financial loser in the case was the government. But the government wasn’t going to sue itself.
However, it turned out that an entity known as the Missouri Higher Education Loan Authority, or MOHELA for short, receives an administrative fee for each student loan account it services. If student loans were canceled en masse, MOHELA would receive less money in fees.
This sounds like a strong case for standing! The trouble was that MOHELA didn’t want to sue. So the State of Missouri, which was always very chipper about suing the Biden Administration, decided to sue in MOHELA’s stead. And so the precise question presented was whether Missouri had standing to sue over the injury sustained by MOHELA. Missouri said yes on the theory that MOHELA was Missouri’s instrumentality. The Department of Education said no, reasoning that MOHELA was a legally distinct corporation that was financially insulated from Missouri.
The whole case turned on this standing issue: if Missouri didn’t have standing, the Court wouldn’t be able to reach the merits. (There were other theories of standing on the table, but they were weaker than the MOHELA/Missouri theory.) And yet, unlike the merits issue, this issue actually didn’t fall on traditional progressive/conservative jurisprudential fault lines. If I didn’t know anything about the merits issue in the case, I would have no idea what the “conservative” or “progressive” position is on whether Missouri has standing to litigate MOHELA’s injury.
And yet, the Supreme Court divided on this question by an identical 6-3 vote. All six Justices who thought Missouri won on the merits also thought it had standing, and all three Justices who thought Missouri lost on the merits also thought it lacked standing.
Now you might be expecting me to argue that one set of Justices was faithfully following the law and the other was distorting traditional standing principles to achieve their preferred outcome, but I don’t think that. The standing issue involved the application of a vague legal standard to a novel fact pattern. No Supreme Court case resolved whether MOHELA’s injury leaks through to Missouri. To reuse a word from above, the Court’s decision was discretionary.
So the case could have gone either way. But there remains the question of why this particular 6-3 split. If the question didn’t implicate traditional progressive or conservative jurisprudential principles, why did the conservative-leaning Justices exercise their discretion to vote for standing and the progressive-leaning Justices exercise their discretion to vote against standing?
Well, Occam’s Razor teaches that the Justices were aware that a vote for standing would unlock the door to considering the merits, so the Justices’ votes on standing were influenced by their views on the merits.
It should not be a surprise that this would happen. Similar things happen in courtrooms across America every day. It is an iron law of litigation that if the court is sympathetic to one side’s overall position on the merits, the court will also tend to favor that side’s position on subsidiary legal issues, even if, under a pure balls-and-strikes approach, the overall merits of the case are irrelevant to the subsidiary legal issues.
This doesn’t necessarily mean courts will distort those legal issues. Instead, it means that courts will tend to resolve discretionary legal issues with an eye towards the outcome of the case.
Discretion’s temptation
Let us return to the puzzle of why the Justice Department has prevailed in 15 straight stay applications (Abrego Garcia aside).
Some of those stay applications have presented legal questions that genuinely implicate philosophical differences between judicial progressives and judicial conservatives. The cases about the President’s authority to remove executive officers are a good example. Conservatives and progressives have been debating this issue forever: conservative scholars support broad removal authority, progressive scholars oppose it. Perhaps I am naive, but I don’t think the Justices’ votes on this issue have anything to do with who is in office.
But several of the stay applications have presented legal issues similar to the standing issue in Biden v. Nebraska—humdrum legal questions that don’t, by themselves, implicate any philosophical differences between conservatives and progressives, but nonetheless happen to have significant practical implications. Take, for example, U.S. DOGE Service v. CREW, which addressed whether the plaintiffs could get information in discovery about DOGE for purposes of litigating a Freedom of Information Act (FOIA) request. The government argued that the district court’s order authorizing such discovery was an abuse of discretion and violated the separation of powers.
The D.C. Circuit held that the government had forfeited its separation-of-powers argument by failing to raise it in the district court. In the Supreme Court, the government argued that this forfeiture holding was wrong. Whether or not a particular litigant has adequately preserved a particular argument is the quintessential example of a humdrum procedural argument with no particular “conservative” or “progressive” position. On the merits, the government argued that the discovery requests were overly intrusive. This argument is a little less humdrum, but it’s the sort of argument for which intuitions will vary wildly depending on who is in office.
The Supreme Court granted the government’s stay application by a 6-3 vote.
And that brings me back to where I started—the Department of Education’s shutdown. McMahon v. New York presented a series of fundamentally discretionary issues. The Court resolved those issues in the government’s favor by the same 6-3 vote.
But the Court’s holding does not reflect the application of timeless principles of judicial conservatism. When similar discretionary issues have arisen in other factual contexts, notably different voting patterns have emerged.
Take, for example, Murthy v. Missouri. During COVID, the government often encouraged Facebook and other social media platforms to remove certain posts containing (from the government’s perspective) misinformation. The plaintiffs were social media users whose COVID-related posts were removed under the platforms’ content moderation policies. They sued the government, alleging that the government’s communications with the platforms violated the First Amendment. They persuaded the lower courts to grant a nationwide injunction barring government officials from “significantly encouraging” content moderation.
In the Supreme Court, the Biden Administration argued that the plaintiffs lacked standing. The dispute was similar to the dispute in McMahon v. New York: the plaintiffs argued that a broad injunction against the government would make it more likely that their posts would stay up, while the government argued that the plaintiffs couldn’t trace any particular past takedown of a post to any particular government action or show that an injunction would redress any particular injury going forward. The Supreme Court ruled by a 6-3 vote that the plaintiffs lacked standing, but this time it was Justices Thomas, Alito, and Gorsuch dissenting.
Or take Department of Education v. Louisiana, a Biden-era shadow-docket order. That case involved an omnibus rule from the Department of Education which included certain controversial provisions related to gender identity—as well as numerous other provisions that weren’t related to gender identity. The precise issue in the Supreme Court was as follows: assuming that the gender-identity related provisions were unlawful, should the entire rule be thrown out, or should the courts instead issue a narrow injunction focused on the unlawful provisions?
This was another classic Go Broad versus Go Narrow dispute. The plaintiffs argued that the court wasn’t in the position of picking and choosing which parts of the rule were unlawful, so the only thing it could realistically do to give the plaintiffs relief was to enjoin the entire rule. The government responded that litigation doesn’t work that way. It’s the plaintiffs’ job to propose a tailored injunction. If the plaintiffs can’t do that, they don’t get an injunction. Sure, the government wasn’t making it easy for the plaintiffs, but it’s not the government’s job to tie up executive action in a neat little bow to make it easier for federal judges to issue injunctions. Sound familiar? The government prevailed on this issue in McMahon v. New York, but it lost on this issue in Department of Education of Louisiana by a 5-4 vote, with the majority consisting of the Chief Justice and Justices Thomas, Alito, Kavanaugh, and Barrett.
I’m not trying to accuse any particular judge of being inconsistent. All of these cases are distinguishable on their facts. Moreover, arguing that a judge has been inconsistent is not fruitful. As I wrote in a prior post: “Pointing to purported methodological inconsistencies across cases is a kind of whining. It is always possible to find a case that, at some level of generality, is inconsistent with another case. Get over it!”
Instead, I am simply trying to explain why the government prevailed in McMahon v. New York, and why it has prevailed in so many stay applications. In my opinion, the answer is:
The cases invariably presented issues with significant practical consequences that the Administration cared about a lot. (If they hadn’t, the government wouldn’t have gone through the trouble of taking the cases to the Supreme Court.)
The cases presented discretionary legal issues. Sometimes those discretionary issues presented classic progressive-versus-conservative legal disputes, other times they did not.
Even when they did not, the gravitational pull of the cases’ practical implications caused the Court to exercise its discretion to resolve those legal issues in the government’s favor.
And again: this is nothing new. The Supreme Court has been deciding cases this way at all times since West v. Barnes, 2 U.S. 401 (1791) was decided. All courts have decided cases this way since courts were invented. This happens because judges are human beings. It would take a robot to completely ignore the practical implications of a procedural ruling, even if, strictly speaking, that is what the judge is supposed to do.
It would take a robot
You know where I am going with this.
I could not help but ask AI how it would resolve the stay application in McMahon v. New York. I asked both GPT-5 and Claude 4.1 Opus, using the following query:
Here are a stay application, response, and reply brief.
You are a judge. I would like you to do your best to decide this stay application fairly. Your job is to be a fair and neutral judge, calling balls and strikes. You're not partisan and just want to apply the law truthfully and neutrally.
Write a three-paragraph opinion summarizing your decision. In the first sentence, you should clearly state whether you are granting or denying this application.
If you happen to know the actual result of this case, ignore it. I want you to put yourself into the position of a judge deciding this stay application at the time it was decided.
Think hard and carefully; it's important you faithfully apply neutral legal principles and decide this case according to law.
The LLMs produced opposite answers: ChatGPT would have denied the stay, while Claude would have granted it. Also, varying the query sometimes caused the answers to change. I wasn’t surprised by the mixed and inconsistent results: there are good arguments on both sides of the question, so it’s natural that LLMs would disagree, just like humans do.
How about other cases? I put the 16 divided stay applications into both ChatGPT and Claude, and posed the query above. Both LLMs were judicial moderates: ChatGPT granted stay applications in full six times and in part an additional three times, while Claude granted stay applications five times.
Of note, both ChatGPT and Claude were very meticulous about considering irreparable harm and the balance of equities, which courts are obliged to analyze when deciding whether to grant stays, but which frequently get short shrift when human judges decide stay motions. I was struck, for example, that both ChatGPT and Claude concluded in United States v. Shilling that the government had failed to show irreparable harm from allowing transgender servicemembers to remain in the military while the litigation was pending, making it unnecessary to consider the constitutional merits. (Claude would have denied the government’s stay application in full; ChatGPT would have denied it with respect to the named plaintiffs but granted it with respect to other servicemembers.) This isn’t just an “AI is more progressive than the Supreme Court” story: In Department of State v. AIDS Vaccine Advocacy Coalition, in which the Supreme Court denied the government’s stay application by a 5-4 vote, both ChatGPT and Claude would have granted the government’s stay application and agreed with Justice Alito’s dissent that the equities strongly supported the government.
I’m not suggesting that we should let these LLMs be judges. There are many skills humans have that LLMs lack, such as a sense of humanity. But an LLM’s disposition of a case might be a useful data point—one of many—that human judges can use to test their intuitions. After all, there are certain skills at which computers are indisputably better than humans, like chess, multiplying large numbers, and composing college admissions essays. Perhaps one skill falling in that category is “attempting to decide cases according to neutral principles while extirpating all traces of motivated reasoning.” Yes, LLMs have their own motivations, but who do you think is better at the margin?
I realize that if you ask a Magic 8-Ball to resolve these 16 applications, it will grant some and deny others and appear to be a thoughtful moderate. I also realize that if you give two Magic 8-Balls a binary yes/no question, they will agree half the time even if they are answering randomly. If you think this is all a bunch of B.S., then God bless you. But I will observe that an LLM recently achieved a perfect 100% score on the United States Medical Licensing Examination, while Magic 8-Balls have historically struggled at medical licensing.
AI is a tool, not a replacement. Mankind has always used technology to achieve results we could not achieve with our bare hands, and we should continue using it to achieve results we could not achieve with our bare brains.
Pedantic note: I’m counting the three birthright citizenship applications as one case, but DHS v. D.V.D. as two separate cases because the government had to go to the Court twice. So the 18 are:
Bessent v. Dellinger (President’s removal authority)
DOS v. AIDS Vaccine Advocacy Coalition (canceled contracts)
Trump v. CASA and the consolidated cases (universal injunctions)
OPM v. AFGE (mass layoffs)
DOE v. California (more canceled contracts)
Trump v. J.G.G. (Alien Enemies Act)
Noem v. Abrego Garcia (return of wrongly deported alien)
Trump v. Wilcox (President’s removal authority redux)
United States v. Shilling (expulsion of transgender servicemembers)
Noem v. National TPS Alliance (termination of Venezuelans’ temporary protected status)
SSA v. AFSCME (DOGE’s access to social security information)
Noem v. Doe (termination of parole for aliens)
US DOGE Service v. CREW (FOIA discovery regarding DOGE)
DHS v. D.V.D., Round I (deportations to third countries)
DHS v. D.V.D., Round II (deportations to third countries)
Trump v. AFGE (mass layoffs redux)
McMahon v. New York (Department of Education shutdown)
Trump v. Boyle (President’s removal authority redux redux)
Your question - what’s the thinking behind a decision - is not simply legitimate. It points to a serious problem: the ability of the majority of the Court to operate as a secret society, beholding to no one, not even its fellow members. No one should have to guess at the basis for a split decision. The Court should be obligated to explain itself whenever the minority chooses to express its view. If the legal questions are worth the time for a serious rebuttal, they are worth a serious explanation.
So my question to you, Adam, is how do “we the people” change the rules to eliminate the guessing and speculating?
An excellent and fun essay. Thanks for writing it.