This document presents two instruments. The first is a federal statute that Congress could enact today under its existing Article I, § 4 authority. The second is a proposed constitutional amendment that would entrench the underlying principle against future statutory reversal and extend it to state legislative districts (which lie outside Congress's Elections Clause reach).
The statute is the primary instrument. The amendment is offered as a secondary recommendation for the entrenchment problem only.
Part I: The Statute
A Bill
To require the use of a publicly-specified, reproducible, non-partisan algorithmic procedure for the drawing of congressional districts in every State following each decennial census; to provide for judicial review of districting plans for compliance with the procedure; and for other purposes.
Be it enacted by the Senate and House of Representatives of the United States of America in Congress assembled,
SECTION 1. SHORT TITLE; TABLE OF CONTENTS.
(a) SHORT TITLE. — This Act may be cited as the "Neutral Districting Act of 2026."
(b) TABLE OF CONTENTS. —
- Sec. 1. Short title; table of contents.
- Sec. 2. Findings and purposes.
- Sec. 3. Definitions.
- Sec. 4. Mandatory algorithmic procedure for congressional districts.
- Sec. 5. The Independent Districting Standards Board; Algorithm Reference Specification.
- Sec. 6. Public reproducibility; ensemble publication; seed disclosure.
- Sec. 7. Judicial review; cause of action; remedies.
- Sec. 8. Effective date; transition; severability.
SEC. 2. FINDINGS AND PURPOSES.
(a) FINDINGS. — The Congress finds the following:
(1) Article I, Section 2 of the Constitution, as construed in Wesberry v. Sanders, 376 U.S. 1 (1964), requires that Representatives be apportioned among congressional districts of substantially equal population.
(2) In Rucho v. Common Cause, 588 U.S. 684 (2019), the Supreme Court held that partisan gerrymandering claims present nonjusticiable political questions in federal court, leaving congressional districting subject only to such limits as Congress affirmatively enacts under the Elections Clause and as state courts may impose under state constitutions.
(3) Peer-reviewed research, including DeFord, Duchin, and Solomon, "Recombination: A Family of Markov Chains for Redistricting," Harvard Data Science Review 3(1) (2021), demonstrates that contemporary algorithmic methods can sample the space of valid contiguous, population-balanced districting plans in a publicly verifiable and reproducible manner.
(4) Chen and Rodden, "Unintentional Gerrymandering: Political Geography and Electoral Bias in Legislatures," Quarterly Journal of Political Science 8(3) (2013), and subsequent literature, demonstrate that neutrally-drawn districting plans do not, and cannot be expected to, produce seat shares that exactly track popular-vote shares; the geographic distribution of partisan voters in the contemporary United States produces structural deviations from proportionality even under plans drawn without any partisan input. The standard against which a plan should be measured is therefore the distribution of plans produced by neutral procedures, not proportional representation.
(5) Public confidence in the integrity of congressional elections is served when districting plans are produced by a procedure whose inputs, algorithm, randomness source, and outputs are simultaneously public, reproducible by any third party, and free of discretionary partisan input.
(b) PURPOSES. — The purposes of this Act are:
(1) to establish a uniform, reproducible procedure for the drawing of congressional districts in every State, removing partisan discretion from the line-drawing process;
(2) to require that the inputs, randomness, and outputs of every congressional districting plan be made public in a form that permits any member of the public to independently verify the plan's compliance with this Act; and
(3) to provide an enforceable federal cause of action by which any aggrieved voter may challenge a plan that fails to comply.
SEC. 3. DEFINITIONS.
In this Act:
(1) ALGORITHM. — The term "algorithm" means a procedure consisting of a Markov chain Monte Carlo sampler over the space of contiguous, population-balanced graph partitions, conforming to the Algorithm Reference Specification published under section 5.
(2) ALGORITHM REFERENCE SPECIFICATION. — The term "Algorithm Reference Specification" means the technical specification published by the Independent Districting Standards Board under section 5(c).
(3) CONTIGUOUS. — A district is "contiguous" when every census block assigned to the district is reachable from every other census block in the district by a path that traverses only census blocks assigned to the same district and that crosses only census-block boundaries that lie within the State.
(4) DECENNIAL CYCLE. — The term "decennial cycle" means the period beginning on the date of release by the Census Bureau of the redistricting data file under Public Law 94-171 for a decennial census and ending on the corresponding date for the next decennial census.
(5) DISTRICTING PLAN. — The term "districting plan" means an assignment, for each State, of every census block in the State to one of the State's congressional districts.
(6) ENSEMBLE. — The term "ensemble" means a collection of districting plans produced by repeated independent runs of the algorithm with distinct random seeds.
(7) POPULATION-BALANCED. — A districting plan is "population-balanced" when no district's population, as measured by the redistricting data file, deviates from the State's average district population by more than one percent.
(8) REPRODUCIBLE. — A districting plan is "reproducible" when, given the plan's published inputs (geographic data, population data, random seed, and Algorithm Reference Specification version), any third party can, using only those inputs and freely available open-source software, regenerate a bit-for-bit identical plan.
(9) STATE. — The term "State" includes each of the several States but does not include any State entitled to only one Representative under the apportionment then in effect.
SEC. 4. MANDATORY ALGORITHMIC PROCEDURE FOR CONGRESSIONAL DISTRICTS.
(a) GENERAL RULE. — Notwithstanding any provision of State law to the contrary, every State shall draw its congressional districts for each decennial cycle by means of the procedure set forth in this section.
(b) PROCEDURE. —
(1) STEP ONE: INPUTS. — Not later than ninety days after the release of the redistricting data file under Public Law 94-171, the State shall publish:
(A) the official census-block geography file for the State, in the form released by the Census Bureau;
(B) the official census-block population file under the redistricting data file;
(C) the random seed selected under paragraph (2).
(2) STEP TWO: SEED SELECTION. — The random seed required by this Act shall be a 256-bit integer derived as follows:
(A) Not earlier than thirty days nor later than fifteen days before the publication required by paragraph (1), the chief election officer of the State shall designate a date and time, falling not less than fifteen days nor more than thirty days after such designation, at which the random seed shall be generated.
(B) On the designated date and time, the chief election officer shall, in the presence of representatives of each political party that received not less than five percent of the popular vote in the State at the most recent presidential election, generate the seed by computing the SHA-256 hash of the concatenation of:
(i) the closing prices, on the most recent business day prior to seed generation, of the thirty stocks then comprising the Dow Jones Industrial Average, in their order of listing on the New York Stock Exchange;
(ii) the winning numbers of every State lottery drawing held on that business day, in chronological order; and
(iii) the headline of the most-recent print edition of The New York Times, The Wall Street Journal, and the State's largest-circulation daily newspaper, in that order.
(C) The chief election officer shall publish, contemporaneously with the generation, both the inputs to the hash function and the resulting seed, in a form that permits any member of the public to independently verify the computation.
(3) STEP THREE: ALGORITHM EXECUTION. — The State shall execute the Algorithm Reference Specification, with the inputs published under paragraph (1) and the seed generated under paragraph (2), to produce a districting plan for the State.
(4) STEP FOUR: PLAN ENROLLMENT. — The plan produced under paragraph (3) shall, upon publication of its inputs and outputs in the form required by section 7, become the official congressional districting plan of the State for the decennial cycle.
(c) NO DISCRETIONARY ADJUSTMENT. — No person may modify, adjust, or override the plan produced under subsection (b)(3). Any State law purporting to authorize such modification, adjustment, or override is preempted by this Act.
(d) NO USE OF PARTISAN OR INCUMBENCY DATA. — The Algorithm Reference Specification shall not consume as input, and no implementation of the algorithm shall consume as input, any of the following:
(1) the partisan registration of any voter;
(2) the residence of any incumbent or candidate;
(3) the historical vote shares of any precinct, county, or other geographic unit; or
(4) any data field whose primary purpose is the prediction of partisan electoral outcomes.
SEC. 5. THE INDEPENDENT DISTRICTING STANDARDS BOARD.
(a) ESTABLISHMENT. — There is established within the Census Bureau, but operating independently of any other component thereof, an Independent Districting Standards Board (in this Act, the "Board").
(b) COMPOSITION. —
(1) MEMBERSHIP. — The Board shall consist of nine members, appointed as follows:
(A) Three members shall be appointed by the President of the United States, by and with the advice and consent of the Senate. Of these, not more than two may be members of the same political party.
(B) Three members shall be appointed jointly by the Speaker of the House of Representatives and the Minority Leader of the House of Representatives. Of these, one shall be selected by the Speaker, one by the Minority Leader, and the third by agreement of both, and not more than two may be members of the same political party.
(C) Three members shall be appointed jointly by the Majority Leader of the Senate and the Minority Leader of the Senate, on the same basis as in subparagraph (B).
(2) QUALIFICATIONS. — Each member shall hold an earned doctorate or its substantial equivalent in mathematics, statistics, computer science, political science, geography, or a related field, and shall have a demonstrated record of peer-reviewed publication in redistricting methodology, computational geometry, or Markov chain theory.
(3) TERMS. — Members shall serve staggered six-year terms. No member may serve more than two terms.
(c) ALGORITHM REFERENCE SPECIFICATION. —
(1) PUBLICATION. — Not later than two years before the release of each decennial redistricting data file, the Board shall publish a complete Algorithm Reference Specification.
(2) CONTENTS. — The Algorithm Reference Specification shall include:
(A) the precise mathematical definition of the Markov chain, including the spanning-tree sampling procedure, the balance-cut criterion, and the acceptance rule;
(B) the precise definition of the contiguity-graph construction from the census-block geography;
(C) the precise definition of the population-balance tolerance, the burn-in length, the polish phase (including any perturb-and-repolish loop that resamples the chain to escape local minima), and the maximum permissible number of independent re-runs from distinct derived seeds, returning the lowest-deviation result;
(D) a graph-theoretic compactness criterion, expressed as a maximum isoperimetric ratio (cross-cut edges divided by the smaller piece's node count) for any accepted balanced cut, together with (i) a deterministic schedule of threshold relaxation across multi-seed retries beginning at a strict default and loosening monotonically only when no balanced cut exists at the prior threshold, and (ii) a partition-level selection rule under which, among multiple retries that all meet population balance, the partition with the lowest mean cross-edge count per district is chosen, preserving ergodicity of the Markov chain while explicitly optimizing for compactness;
(E) a complete reference implementation in a publicly-readable programming language, released under a permissive open-source license;
(F) a corpus of test inputs and expected outputs sufficient to verify that any independent implementation produces bit-for-bit identical results; and
(G) the deterministic pseudorandom number generator to be seeded by the random seed under section 4(b)(2).
(3) PEER REVIEW. — Before publication under paragraph (1), the Algorithm Reference Specification shall be subjected to public peer review, with a comment period of not less than one hundred eighty days, and the Board shall publish a written response to every substantive comment received.
(4) AMENDMENT. — The Board may amend the Algorithm Reference Specification between decennial cycles only by the same peer-review procedure required under paragraph (3), and any such amendment shall take effect only at the next decennial cycle following its publication.
(d) PROHIBITION ON PARTISAN INPUT. — In developing and amending the Algorithm Reference Specification, the Board shall not consider the predicted partisan effect of any methodological choice, and shall publish, with each version of the Specification, a written certification by every member that no such consideration was made.
SEC. 6. PUBLIC REPRODUCIBILITY; ENSEMBLE PUBLICATION; SEED DISCLOSURE.
(a) REQUIRED PUBLICATIONS. — Within fifteen days after the production of a plan under section 4, the State shall publish, in a single open-data repository hosted by the Census Bureau:
(1) every input file consumed by the algorithm;
(2) the random seed and its full derivation under section 4(b)(2);
(3) the version of the Algorithm Reference Specification used;
(4) the complete output of the algorithm, including the final districting plan and every intermediate state of the Markov chain, in a format sufficient to permit bit-for-bit reproduction; and
(5) a SHA-256 hash chain over all of the above, to permit detection of post-publication modification.
(b) CITIZEN VERIFICATION. — The Census Bureau shall maintain a public verification service that, given any State's published artifacts under subsection (a), independently re-executes the Algorithm Reference Specification and certifies whether the published plan matches the re-computed plan.
(c) PROHIBITION ON SUPPRESSION. — No State, and no officer of any State, shall withhold, redact, or delay the publication required by subsection (a). Any failure to comply shall constitute a violation of this Act.
SEC. 7. JUDICIAL REVIEW; CAUSE OF ACTION; REMEDIES.
(a) CAUSE OF ACTION. — Any individual registered to vote in any State whose congressional districting plan was produced under this Act may bring a civil action in the United States district court for the district in which the individual resides to enforce this Act.
(b) GROUNDS. — A plan may be challenged on any of the following grounds:
(1) The plan was not produced by execution of the Algorithm Reference Specification with the published seed.
(2) The seed was generated other than as required by section 4(b)(2).
(3) The State failed to publish the materials required by section 6.
(4) The plan was modified, adjusted, or overridden in violation of section 4(c).
(c) STANDARD OF REVIEW. — In an action brought under this section, the burden is on the State to establish, by a preponderance of the evidence, that the published plan is the bit-for-bit output of executing the published Algorithm Reference Specification with the published seed on the published inputs. A challenger who presents an independent re-execution producing a non-matching output shall be entitled to judgment as a matter of law unless the State produces such a re-execution of its own.
(d) REMEDIES. —
(1) DEFAULT. — Upon a finding of violation, the court shall enter an order directing the State to re-execute the Algorithm Reference Specification on the published inputs with the published seed, and shall enter the resulting plan as the State's congressional districting plan for the decennial cycle.
(2) REPLACEMENT SEED. — Where the violation consists of the seed itself having been improperly generated, the court shall order the generation of a replacement seed under section 4(b)(2), with the designated date set by the court, and direct re-execution accordingly.
(e) JUDICIAL ECONOMY. — Actions under this section shall proceed in the form of an action for declaratory relief and shall be heard by a three-judge district court convened under section 2284 of title 28, United States Code, with direct appeal to the Supreme Court.
SEC. 8. EFFECTIVE DATE; TRANSITION; SEVERABILITY.
(a) EFFECTIVE DATE. — This Act shall take effect upon enactment, and shall first apply to the decennial cycle following the 2030 census.
(b) TRANSITION. — Until the cycle described in subsection (a), the districting plans in effect on the date of enactment shall continue in effect, except that any State whose plan is invalidated under State or Federal law during the interim period shall produce its replacement plan by the procedure of this Act.
(c) NO RETROACTIVE EFFECT. — Nothing in this Act creates any cause of action with respect to any election held before the effective date in subsection (a).
(d) SEVERABILITY. — If any provision of this Act, or the application of any provision to any person or circumstance, is held to be unconstitutional or otherwise invalid, the remaining provisions and applications shall not be affected.
Part II: A Proposed Constitutional Amendment
Statement of Need
The statute in Part I is enforceable today under Congress's existing Article I, § 4 authority over the manner of holding congressional elections. It does not, however, reach state legislative districts, which are governed by state constitutions and lie outside Congress's reach under the Elections Clause. It is also subject to repeal or modification by any future Congress.
If the goal is to constitutionalize the principle and to extend it to all legislative districting in the United States — federal, state, and local — a constitutional amendment is required.
The proposed amendment is brief, by design. It establishes the principle in a form that admits of legislative implementation without freezing the specific algorithmic details of any particular era's mathematics into the constitutional text.
Article XXVIII (Proposed)
SECTION 1. Every legislative district for the election of any member of the House of Representatives, of any State legislature, or of any local legislative body, shall be drawn by a procedure that:
(1) is publicly specified in advance of its application;
(2) is reproducible by any third party from publicly-disclosed inputs and randomness; and
(3) does not consume as input, and does not condition its output on, the partisan registration of any voter, the residence of any incumbent or candidate, or the predicted partisan effect of any line placement.
SECTION 2. The Congress shall have power to enforce this article by appropriate legislation, and shall publish, in advance of each decennial cycle, the procedure required by section 1 for districts within its Article I, § 4 authority.
SECTION 3. Each State shall publish, in advance of each decennial cycle, the procedure required by section 1 for districts within its authority.
SECTION 4. The judicial power of the United States shall extend to controversies arising under this article. The first sentence of clause 1 of section 2 of Article III, and the second clause of the Eleventh Amendment, shall not be construed to bar any such controversy.
Drafter's Note on Design Choices
A few choices in the foregoing warrant a brief explanation.
Why a Markov chain procedure rather than an optimization procedure. The statute deliberately requires a stochastic sampling procedure (ReCom) rather than a deterministic optimization (e.g., minimize total perimeter). A deterministic optimum is a single map; a stochastic sample is a distribution. The latter makes it possible for any future analyst to characterize the space of valid plans, not merely to verify that one plan was produced correctly. This matters for evaluating whether any particular feature of a plan (e.g., its racial composition, or its partisan tilt under a given electorate) is a typical feature of neutral maps or an outlier. See DeFord, Duchin, and Solomon, supra, on the mixing properties and applied verification advantages of ReCom.
Why a graph-isoperimetric compactness criterion and no other shape-based criterion. The statute imposes contiguity, population balance, and a single shape-based requirement: a graph-isoperimetric compactness threshold on every accepted spanning-tree cut, with a deterministic relaxation schedule and a partition-level selection rule that prefers more compact partitions among those that hit population balance (Sec. 5(c)(2)(D)). That criterion is included because it is mechanical — it requires no human judgment to apply. An algorithm either produces a cut whose cross-edge ratio falls below the published threshold, or it doesn't; a citizen with a laptop can verify the answer.
By contrast, the statute does NOT require county-line preservation, "communities of interest" preservation, geographic compactness measured by continuous-geometry metrics (Polsby–Popper, Reock), or any other criterion whose application depends on a drafter, board member, or judge weighing competing interpretations. Every such criterion is a vector for discretionary application, which is the entrenchment vector for gerrymandering. This is the principal lesson of the post-Rucho state litigation: criteria that require human judgment are criteria that admit human bias. The graph-isoperimetric criterion was chosen precisely because it admits no such judgment — it is a single arithmetic inequality whose threshold is set by the Board, in advance, by peer-reviewed amendment.
Why the seed protocol. A reproducible procedure requires a published seed; a seed that is chosen by a partisan actor permits cherry-picking among many possible maps until a partisan-favorable one appears. The protocol in section 4(b)(2) is designed to make seed cherry-picking infeasible: the seed is generated from inputs (stock prices, lottery numbers, newspaper headlines) that no actor controls in advance, and is published contemporaneously with its inputs so that any modification of the inputs after the fact is detectable.
Why the Independent Districting Standards Board houses the algorithm, not Congress. The mathematics of Markov chain redistricting will improve over the coming decades. Freezing one specific procedure into the United States Code would lock the Nation into 2026's state of the art. The Board, with its peer-reviewed amendment procedure, allows the algorithm to evolve while preserving the political-neutrality and reproducibility properties that this Act requires. The Board's structure mirrors that of the Federal Election Commission and the Federal Reserve Board of Governors in its balanced-appointment structure, and that of the National Institute of Standards and Technology in its substantive-expertise requirement.
Why bit-for-bit reproducibility. The statute requires bit-for-bit reproduction (Sec. 3(8) defines "reproducible"; Sec. 7(c) makes a citizen's independent re-execution producing a non-matching output dispositive in court). This is stronger than "approximately reproduces" or "produces a plan with the same statistical properties." The reason is enforcement: any deviation from bit-for-bit identity is detectable by a citizen with a laptop, and the cause-of-action structure in Sec. 7 makes such deviations grounds for invalidation. Weaker reproducibility standards would shift the enforcement burden from arithmetic verification to expert-witness statistical contestation, which is precisely the discretionary vector this Act seeks to eliminate.
Why neutral plans aren't proportional, and why that's the right benchmark. The Findings in Sec. 2(a)(4) acknowledge explicitly that algorithmically-neutral plans do not produce proportional seat shares. This is essential. A statute requiring proportional outcomes would require building in a partisan correction to counteract the structural bias that geography produces — which is itself a form of gerrymandering, just one in the opposite direction. The honest standard is: a plan should look like a typical plan from the neutral-procedure distribution. Plans that lie far in the tail of that distribution are suspect; plans that lie near the median are not. The dashboard's contemporary outputs make this concrete: in a year like 2024 with a 49.2 D / 50.8 R two-party popular vote, the algorithmically-drawn House lands at roughly D 197 / R 238 — a 2-3 point structural R advantage in seats arising from the geographic distribution of partisans, not from any drawing choice. Reproducing that structural advantage in a neutral plan is a feature, not a bug, of the procedure. See Goedert, "Gerrymandering or Geography? How Democrats Won the Popular Vote but Lost the Congress in 2012," Research and Politics 1(1) (2014); Chen & Rodden, "Unintentional Gerrymandering," Quarterly Journal of Political Science 8(3) (2013), discussing the structural origins of this asymmetry.