Friday’s post argued that what the Court did in Louisiana v. Callais is best described as breaking a structural link — a load-bearing connection that was holding up many actions at once, not just the Louisiana map. That post was about what. This post is about why now. Why this case, why this term, why — for that matter — thirteen years after Shelby County v. Holder, eleven years after Allen v. Milligan, and only one year after the Court initially heard Callais arguments and declined to decide it.
The case was argued on March 24, 2025. It was not decided that term. In June 2025, the Court issued an order restoring it to the calendar for reargument and directing the parties to brief a question broader than the one originally presented. The original question asked whether Louisiana’s SB8 was an impermissible racial gerrymander. The supplemental question asked whether the State’s intentional creation of a second majority-minority district violates the Fourteenth or Fifteenth Amendments — a question that does not turn on the specific facts of SB8 at all, and whose resolution would, almost necessarily, address the broader constitutional architecture of Section 2 districting. Reargument took place in October 2025. The decision came down last Wednesday.
None of this is in itself unusual. Cases get held over. Reargument orders happen. Supplemental briefing is requested when a Court wants more before it commits. What is worth noting is the shape of what the Court did during the year it had Callais in hand. It did not simply delay. It reformulated the question. The case that was eventually decided is, in a doctrinal sense, a different case from the one that was argued in March 2025. The scope expanded. The set of issues the eventual opinion would resolve grew. By the time the decision arrived, the Court was not answering a question about Louisiana. It was answering a question about Section 2.
In 2017, with Deborah Beim and Tom Clark, I published a paper called “Why Do Courts Delay?”1 The setup is simple. A court has a docket of disputes, each of which can be resolved or left alone. Resolving a dispute costs something; leaving it costs something. But disputes also have linkages: resolving one dispute can provoke others — the children, the grandchildren, the questions that follow logically once the first is answered. A forward-looking court does not just compare the cost of resolving today against the cost of waiting. It also weighs the cost of the resulting progeny — the new disputes that resolution will bring onto the docket tomorrow.
Two predictions of the model are worth keeping in view. First, a court will sometimes rationally delay resolving a dispute it could easily resolve, because the downstream consequences of resolution are too costly to absorb on their own. Second — and this is the prediction that does the work today — when several disputes share progeny, the court has an incentive to resolve them together. Bundled resolution lets the court absorb the downstream consequences once instead of repeatedly. The model calls this an endogenous economy of scale: not because resolving cases gets cheaper as you do more of them in some mechanical sense, but because resolving cases that share offspring lets you handle the offspring as a single bundle rather than a series of recurring problems.
The case study in the paper is the use of statistical evidence in litigation. The Supreme Court declined to articulate any doctrine on statistical evidence for nearly a century after the question first arose, despite obvious accumulating costs from the absence of doctrine. Then, in 1977, the Court resolved three discrimination cases in a single term — Castaneda v. Partida (jury selection), Teamsters v. United States (private employment), and Hazelwood School District v. United States (public employment) — using each as a vehicle to articulate principles that applied across all three. The Court had not become more interested in statistical evidence in 1977. The Court had finally accumulated enough logically-connected disputes that bundled resolution was efficient. The progeny of any one case — the secondary questions about expert standards, admissibility, discovery — would have been costly to handle alone. Three cases sharing those progeny made the bundle worth resolving.
The Callais docket has the same shape. Held over for a year. Returned with a broader question. Decided alongside (and effectively pre-disposing of) a number of pending lower-court §2 challenges that share the progeny the Louisiana case was already going to raise. The Court did not just delay. It used the year to bundle.
And here the Statistical Evidence comparison earns its place, because the parallel is not just structural. It is also instructive about something the formal model does not, on its own, address.
In 1977, the Court delayed and bundled in order to open a doctrine. The trilogy admitted statistical evidence into discrimination litigation, generated a new line of cases, and structured the field for the next several decades. The downstream consequences were extensive but manageable, and the doctrine grew out of the bundle. In 2026, the Court delayed and bundled in order to close a doctrine. The post-Callais downstream consequences will not extend the doctrine. They will prune it. The same operator, run with opposite normative valence.
This is a recurring pattern in formal-theory analyses of institutions, and it shows up in the algorithmic-classification work I have been doing with Maggie Penn as well. The mathematics of a procedure does not, on its own, tell you what the procedure is for. A test that maximizes accuracy can be deployed to expand who is admitted or to restrict who is admitted, depending on what one means by accuracy and accuracy for whom. The Beim-Clark-Patty operator, applied in 1977 and again in 2026, gives us the same prediction in both cases: when shared progeny accumulate, expect bundled resolution. What the model does not tell us is whether the bundle will be used to extend the doctrine’s reach or to retract it. That is a question about the goals of the actor running the operator, not about the operator itself.
Almost every formal apparatus in social science has this property. The model predicts the operation. The interpretation supplies the direction. People who are uncomfortable with this fact tend to call it a limitation. I think it is a feature: it forces the analyst to be explicit about what the operator is being used for, rather than letting the formalism quietly carry the normative weight.
So far the story is: Callais took thirteen months because the Court was bundling. But that does not explain the larger thirteen-year arc — Shelby County in 2013, Brnovich in 2021, the failed-then-revisited Allen sequence, and now Callais. The within-case story (delay-and-bundle) does not, on its own, explain the across-case sequence. To make sense of the longer arc, the apparatus needs an addition.
Here is what I think the addition is. The Why Do Courts Delay? model treats the structure of doctrinal links — which arguments support which actions, and at what cost — as given. Friday’s post, drawing on my 2008 paper on arguments-based collective choice, treats the Court’s link-breaking operation as the response to a known cost structure: the Court breaks a link when the cumulative gain from overturning all the actions the link supports exceeds the cost of breaking it. Both apparatuses assume the Court knows what the costs are.
I do not think the Court does know. I think the Court has been finding out.
Read Shelby County, Brnovich, and the smaller cases between them as informative trials. Each one applies stress to a different part of the Voting Rights Act’s structural support and observes how the surrounding system responds — how lower courts absorb the new doctrine, whether Congress reacts, whether state legislatures behave as predicted, whether the public registers the change as significant or routine. Shelby County struck down the preclearance coverage formula in 2013 and produced a measurable but bounded political response. Brnovich in 2021 narrowed Section 2’s reach in voting-procedure cases and produced less response than Shelby. Allen v. Milligan in 2023 declined an opportunity to break the link Callais would later break, in a context where (we now know) the Court was not yet ready to act on the central link. Each decision returned information about how much weight the surrounding scaffolding could absorb. Each decision tightened the Court’s posterior on the cost structure of the doctrine it was working through.
By 2026, the posterior was tight enough to act on at the load-bearing center. Callais is the operation the prior thirteen years of measurements made possible.
This reading is consistent with what observers have been describing all along. CNN’s Joan Biskupic noted last weekend that the Roberts Court’s count of explicit overrulings is misleadingly low because the Court has been “pummeling” precedent without explicitly killing it. That is a precise observation. Pummeling, in the formal sense I am proposing, is informative stress-testing: each blow returns data on how much the structure can absorb without snapping back. The Court’s count of formal overrulings is low because the Court has been measuring, not yet acting. Callais is the action the measurements were preparing for.
Three operators, then. The arguments-based-choice operator identifies which structural link, broken, will produce the largest doctrinal cascade. The delay-and-bundle operator identifies the moment at which enough logically-connected disputes have accumulated that bundled resolution is efficient. The stress-testing operator runs informative trials — cases that probe the surrounding structure without committing the central operation — to learn the cost structure on which the first two operators depend. Composed in sequence, they describe what a Court does when it is engaged in the sustained reshaping of a body of doctrine over more than a decade. The reshaping is not erratic. The pace is not random. The cases that look in isolation like inconsistency or hesitation look, under the composed apparatus, like a coherent program of measurement followed by bundled action.2
I do not think this composition is the Roberts Court’s invention. The pattern likely shows up in other multi-decade doctrinal projects — the building of administrative-state doctrine across the Burger and Rehnquist Courts, perhaps, or the slow construction of the modern First Amendment from the 1960s onward. In every case where a Court is reshaping a body of doctrine over a period long enough that the doctrine’s cost structure is itself uncertain, you would expect to see something like the composition: measurement, bundle, action. The composition is not unique to dismantling. Doctrines have been built this way too. What is unusual about the present moment is that the apparatus is visible. The cases are recent enough, the actors candid enough, and the news coverage attentive enough that the operation can be described while it is still happening. Most of the time, this kind of analysis becomes available only in retrospect.
There is a third post in this series, which I will write in two weeks once the redistricting story has had time to develop. The argument so far has been about the Court. The next argument is about what happens after the Court acts. State legislatures are now redrawing maps in real time — Florida passed one within an hour of the Callais decision; Mississippi has called a special session; Tennessee’s governor has announced a target. This is the iterated game the Court has just unlocked, and it has the structure of an arms race in which there is no Schelling point to coordinate on. That is the next post.
For now, the point I want to leave you with is the one the formal apparatus, taken together, makes available. When you read commentary describing the Court as inconsistent, or as moving slowly, or as occasionally restrained and occasionally aggressive, ask whether the commentator is describing the operation or the appearance of the operation. Inconsistency at the level of individual cases is consistent with coherence at the level of the program. Slow movement on a given doctrine is consistent with rapid measurement of the surrounding structure. The cases that did not seem to do much at the time may have been the most informative. Allen is now visibly such a case. So, in retrospect, was Brnovich. So, very visibly, was Shelby.
The thirteen years were not hesitation. They were calibration.
With that, I leave you with this.
Notes
1 Deborah Beim, Tom S. Clark, and John W. Patty, “Why Do Courts Delay?” Journal of Law and Courts 5(2): 199–241 (2017). The model is decision-theoretic: a single forward-looking court chooses which of a set of disputes to resolve, knowing that resolution provokes successor disputes according to a fixed linkage matrix. The two predictions emphasized here — rational delay and bundled resolution — are propositions 1 and 2 of the paper.
2 The stress-testing extension I sketch here is a natural reading of what happens when one relaxes the Beim-Clark-Patty assumption that the linkage matrix and its costs are common knowledge. Once those costs are uncertain, sequential adjudication becomes a sequential experimentation problem, and earlier cases acquire informational value above and beyond the actions they directly resolve. The full formal treatment of this is a paper, not a blog post; this is a sketch, with the expectation that it will be developed properly elsewhere.