Algorithmic Redistricting: Prove the Map, Don't Just Draw It

The current landscape: mid-decade redistricting without a federal backstop

Since mid-2025, several states have redrawn congressional maps outside the normal once-a-decade cycle, the largest coordinated mid-decade push in modern history. Texas Republicans redrew their map in 2025 to protect the party's narrow U.S. House majority; a three-judge federal panel found the state had likely engaged in racial gerrymandering targeting five Black and Latino members' seats and blocked the map. The Supreme Court first stayed that block in December 2025 (Abbott v. LULAC, No. 25A608), then went further on April 27, 2026, summarily reversing the district court's judgment outright over a dissent from Justices Sotomayor, Kagan, and Jackson (Abbott v. LULAC, No. 25-845). California answered with Proposition 50, a Democratic-backed ballot measure voters approved in November 2025 that redraws the state's map through 2030; a three-judge federal panel rejected a Republican racial-gerrymandering challenge 2-1 on January 14, 2026, and the Supreme Court denied an emergency injunction pending appeal in February 2026. Missouri's Republican legislature, in a special session Governor Mike Kehoe called in August 2025, redrew the Kansas City-area seat held by Democratic Rep. Emanuel Cleaver into a more Republican-leaning district; the Missouri Supreme Court upheld the map 4-3 in March 2026. Ohio, the one state legally required to redraw its map before 2026 under a 2018 constitutional amendment, produced the year's most bipartisan outcome: after the legislature missed its deadline, the Ohio Redistricting Commission unanimously approved a new map on October 31, 2025.

Four states, four outcomes, one shared feature: the map that mattered was the one a legislature or a ballot measure produced, and the fight that followed was almost entirely about whether a court would let it stand, not about what evidence anyone had to show before adopting it.

Two rulings that define what's left

Rucho v. Common Cause held that partisan-gerrymandering claims are a political question federal courts cannot hear at all, a 5-4 ruling from 2019 that remains controlling. That left two paths open: state courts interpreting their own constitutions (most state courts that have taken up the question have found limits there, even where the federal one has none), and federal claims framed around race rather than partisanship, chiefly racial gerrymandering and Voting Rights Act Section 2 claims. On April 29, 2026, Louisiana v. Callais narrowed the second path substantially. The Court's 6-3 ruling (Alito writing, Thomas and Gorsuch concurring, Kagan, Sotomayor, and Jackson dissenting) reworked the framework Section 2 claims have used since Thornburg v. Gingles (1986) along three axes. A plaintiff's illustrative map, offered to show an additional majority-minority district was achievable, must now be built without race as a districting criterion and must independently satisfy the state's own legitimate districting goals, including partisan ones the state has specified. Racial-bloc-voting evidence must control for party affiliation, showing a divide plain partisan voting patterns don't already explain. And the totality-of-circumstances inquiry must focus on present-day intentional discrimination, not a state's discriminatory history. The Court held that Louisiana's own second Black-majority district failed that standard, since Section 2 didn't actually require it, and so using race to draw it was itself an unconstitutional racial gerrymander. Justice Kagan's dissent characterized the new framework as reviving an intent-based standard Congress repudiated in 1982 and estimated the ruling would eliminate most Section 2 claims going forward; the majority disputed that characterization. Combined with Rucho, the current federal landscape places partisan gerrymandering fully outside federal courts' reach and subjects Section 2 vote-dilution claims, and a state's ability to use Section 2 compliance to justify race-conscious districting, to a considerably higher bar than they faced a year earlier.

What "algorithmic redistricting" already means in practice

No state, and no serious academic project, has proposed letting software draw a legally binding map with no human sign-off. What already exists, and has existed since at least 2021, is narrower: an algorithm generates a large sample of alternative maps that follow the same rules (population equality, contiguity, compactness, sometimes county or community-of-interest preservation) the adopted map is supposed to follow, then measures whether the adopted map is a statistical outlier against that sample. MGGG, the Tufts research group behind the open-source GerryChain library and its "recombination" sampling method, doesn't run elections; it produces the ensemble that outside evaluators, expert witnesses, and academic reviewers of real commission and legislative cycles use as an input to their own analysis. Princeton's Gerrymandering Project and RepresentUs publish the same kind of analysis publicly for every state through the Redistricting Report Card, simulating roughly a million plans per state; it gave Texas's 2021 map an F for partisan and racial-dilution concerns, and handed the same grade that cycle to Illinois, Maryland, New York, Ohio, and Oregon, states controlled by both parties. The Institute for Computational Redistricting at the University of Illinois worked directly with Arizona's independent commission and prepared analytical plans for Missouri redistricting advocates, describing its own role as producing options "for consideration," not decisions. Every working example points the same direction: evidence and options in, a human body's judgment out.

The neutrality problem

A recurring claim, appealing enough that a Duke Law Journal piece carries almost this exact title, is that a computer algorithm can draw district lines without bias. That claim doesn't survive contact with what the algorithm has to be told: which criteria to satisfy, how heavily to weight compactness against competitiveness, and, most consequentially, whether to use racial and ethnic population data at all. Compactness sounds neutral and isn't applied evenly, since a party whose voters cluster densely in cities is affected differently by a strict compactness rule than one whose voters are spread out, a contested question in the academic literature rather than a settled one. The sharper problem is what happens when a designer excludes race entirely on the theory that "race-blind" means neutral. Moon Duchin and Douglas Spencer tested that theory directly: an ensemble built on race-blind inputs substantially undercounted Black-majority districts compared with maps drawn using racial data. Excluding race isn't a neutral default. It's a design choice with a measured, unequal outcome, and it carries more weight now that Callais has narrowed the legal backstop that used to catch that outcome after the fact.

Three ensembles with distinct jobs

Proposals 1, 2, and 4 sound like they could be the same tool used three ways. They can't be, and collapsing them into one would break the mechanism. Proposal 1's public baseline has to stay strictly neutral, traditional criteria only, because its whole job is catching an outlier: if a state could feed its own declared political goals into the baseline it's measured against, the baseline would simply absorb whatever partisan target the state announced, and the map would stop looking like an outlier no matter how skewed it actually was. A baseline that can be told what to expect isn't a baseline. Proposal 2 exists because the neutral baseline has its own known blind spot: Moon Duchin and Douglas Spencer's research found that excluding racial data doesn't produce a fairer comparison, it produces one that systematically undercounts minority-opportunity districts, so a second, race-aware audit runs alongside the neutral one specifically to catch what the neutral version misses. Proposal 4 answers a narrower, later question: once a Section 2 plaintiff is actually in federal court, Louisiana v. Callais (2026) requires an illustrative map used as evidence to be built without race as a criterion and to match the state's own stated political goals, not just its traditional map-drawing rules. That construction is done case by case, for litigation specifically, and it never substitutes for the standing public baseline in Proposal 1, which has to keep working before any lawsuit exists.

Proposal mechanics and objections addressed

The core mechanism ties to Congress's Elections Clause authority over the times, places, and manner of congressional elections, so this proposal applies only to congressional maps, not state legislative ones, to stay on the clearest constitutional ground. It doesn't require any state to adopt an independent commission; a legislature that draws its own map can still comply by publishing the ensemble comparison before final adoption. Two objections surface immediately. First, federalism: legislators in Missouri and Ohio have both resisted federal involvement in redistricting procedure in the current cycle. The answer is that a disclosure requirement regulates process, not outcome, and leaves each state's chosen map-drawing body fully intact; Congress has long used its Elections Clause authority to set procedural conditions on congressional elections without dictating who wins them. Second, why settle for disclosure instead of directly barring maps that fail the ensemble test: not because Rucho forecloses it. It explicitly does not; the majority itself says statutes and state constitutions can supply standards for courts to apply, and names congressional action under the Elections Clause as an available remedy. A federal statute banning outlier maps is constitutionally open post-Rucho. This proposal chooses disclosure anyway as a deliberate design choice, not a constitutional workaround: writing a workable statutory line, how much deviation from a criteria-compliant baseline is illegal, without either being too loose to bind or too rigid to survive contact with a state's legitimate map-drawing choices, is still a hard problem Congress hasn't attempted and this proposal doesn't ask it to solve first. Disclosure sidesteps that unsolved design problem entirely: it doesn't ask a court to decide how much skew is too much, only to confirm the required analysis was published. This proposal accepts a trade-off: it makes gerrymandering harder to do quietly, not impossible to do openly. A legislature with full political control can still adopt an outlier map and accept the public cost of having said so in the open. That is a narrower promise than banning gerrymandering outright, and it is the more achievable one while an outright statutory ban's own design question remains unsettled.