Methodology.

The Engine module of the Lab generates plain-language summaries, legal-effect analyses, distributional-impact narratives, and tier-classification reasoning for every bill the Lab covers. The Engine is, in part, an AI analytical pipeline. The standard the Lab holds the Engine to is full reproducibility.

What the Engine does

For each bill, the Engine produces:

• Plain-language summary. What the bill does, in citizen-readable English. Drawn from the bill's text and contextualized against current doctrine.
• Legal-effect analysis. What the bill changes in current law, what authorities it invokes, what doctrinal questions it raises.
• Distributional-impact narrative. Who is affected by the bill, in which directions, with what magnitude where measurable.
• Tier-classification reasoning. Why the bill is classified as Tier 1, 2, or 3, with explicit reasoning and citation.
• Manuscript anchor. What chapter and section in The First Three Words the bill maps to, with line range.

Reproducibility protocol

Every Engine analysis is produced by a documented pipeline. The pipeline is reproducible — meaning a sufficiently equipped reader could re-run the analysis on the same inputs and get a comparable output. The Lab guarantees the following per analysis:

• Model version. Recorded at generation time. Visible on the bill card or on a "Show the prompt" affordance attached to it.
• Prompt. The exact prompt used to generate each analytical section. Published. Versioned.
• Run artifact. The raw output of the model run, retained with timestamp and prompt hash. Not necessarily published in full (length); available on request and via the GitHub repository where the artifacts live.
• Editorial review record. Every Engine output is reviewed by an editor (currently the editor; over time, by a domain panel). The review either accepts the output as-is, accepts with revisions (revisions logged), or rejects it (logged with reasoning, and the analysis is regenerated against a revised prompt).

Why this matters

The civic-tech graveyard is full of platforms that produced AI analytical output without provenance. When a model update later contradicted an early analysis, the platform had no defense — readers could not see what generated what. The Lab's reproducibility protocol is the defense. A reader who finds an analysis they disagree with can: see the prompt, see the model version, re-run the analysis on their own infrastructure, and file a dispute if the regenerated output differs. The dispute process (see /contribute#tier-classification) handles the resolution.

Current state

The Engine ships live on a seeded corpus of five federal bills selected to span the three reform tiers: a Tier 1 procedural-transparency bill (Shadow Docket Sunlight Act, S. 3533); a Tier 2 reform of the Blueprint's central illustrative reauthorization (NFIP Reauthorization and Reform Act, H.R. 5484); a Tier 2 administrative-modernization measure (Prescription Information Modernization Act, H.R. 4132); a Tier 2 Mission-Domain straddle (Thomas R. Carper Water Resources Development Act, S. 4367); and a Tier 2 fiscal-architecture statute (Fiscal Commission Act, H.R. 3289). Every bill's detail page renders the four analytical panels (Plain-Language Summary; Amendment Tracking; Legal-Effect Analysis; Distributional Impact) plus the Tier Classification & Pathway panel, each accompanied by the methodology callout that documents data sources, model version, run ID, the Lucas Critique caveat on forward-looking projections, retrieval date, and a content hash for the bill text the analysis ran against.

Phase 2 expands the seeded corpus from five bills to ten to fifteen, introduces live API integration for new bills as they are introduced and amended, and adds the cross-module hand-off so an Engine output can be loaded directly into the Map (for pathway analysis) and the Simulator (for fiscal or districting modeling). The reproducibility protocol above is the standard every Engine analysis is held to — in MVP and in subsequent phases.

The Lab classifies every reform, bill, and Forge artifact under one of three tiers. The classification is consequential — it shapes how readers understand whether reform is reachable in the current cycle or requires constitutional repair. The classification is also editorial; it is not a mathematical assignment.

The three tiers

Tier 1 — Begin tomorrow. State ballot initiatives, municipal or administrative action, federal legislation requiring no new authority. The defining test: can the work begin under current legal authority, in the current legislative or administrative cycle, without requiring new federal statute or constitutional change? If yes, Tier 1.

Tier 2 — Federal statute, testable now. Federal legislation and regulatory architecture that can be designed and tested while the current system operates. Requires new federal statute but not constitutional amendment. The defining test: does the reform require an act of Congress that does not exceed Congress's existing constitutional authority? If yes, Tier 2.

Tier 3 — Requires Article V. Constitutional amendment. The defining test: is the reform foreclosed by current constitutional doctrine, such that no amount of statute or executive action can deliver it without violating the Constitution as the courts currently read it? If yes, Tier 3.

Tier-classification reasoning

Every tier classification published on the Lab is accompanied by reasoning that documents:

• Constitutional authority. What constitutional provision authorizes the reform (or, for Tier 3, what constitutional provision currently forecloses the reform).
• Doctrinal posture. Whether current Supreme Court doctrine permits, requires, or forecloses the reform pathway.
• Legislative or administrative posture. Whether the reform is currently before Congress, in committee, in regulatory rulemaking, in a state ballot process, or only in concept.
• Manuscript citation. What chapter and section in The First Three Words the tier reasoning anchors to.
• Editorial reviewer. Who classified it and when.

Disputes

Tier classifications are open to dispute. Anyone can file a dispute on any reform, bill, or Forge artifact — the process is documented at /contribute#tier-classification. Dispute outcomes are published with reasoning and become part of the Lab's tier-classifications audit trail.

Why this matters

A Tier 1 reform that turns out to require new federal authority misleads citizens about whether action can begin tomorrow. A Tier 3 reform that is actually reachable through statute misleads citizens about whether constitutional change is necessary. The Lab's editorial credibility depends on getting these classifications right and being transparent about the reasoning when it gets them wrong.

Constitutional doctrine moves. Sometimes it moves dramatically — as it did on April 29, 2026, in Louisiana v. Callais, when the Supreme Court inverted the §2 Voting Rights Act protection that Allen v. Milligan (2023) had appeared to preserve. The Lab's redistricting reform card had shipped four days before that ruling; within forty-eight hours of the ruling the Lab's doctrinal notice, bill analyses, and Forge artifact text were updated to reflect the new doctrine.

This page documents the protocol the Lab follows when doctrine changes.

Triggering events

A doctrinal-change protocol review is triggered when:

• The Supreme Court rules on a case the Lab cites as live authority for any reform, bill, or Forge artifact.
• A federal circuit court of appeals issues a ruling that creates a circuit split on a doctrine the Lab cites.
• Congress enacts a statute that materially modifies a regulatory regime the Lab analyzes.
• A state supreme court issues a ruling that affects a Tier 1 state-pathway analysis.

The review

Within seventy-two hours of a triggering event, the editor (and, for major shifts, the doctrinal panel — same composition as the tier-classification dispute panel) conducts a review covering:

• Affected surfaces. Which reform cards, bill analyses, Forge artifacts, scoreboards, and doctrinal notes cite the changed doctrine.
• Required updates. What text, classification, or framing must change.
• Notice requirement. Whether a doctrinal notice (a dismissible top-of-card alert) is warranted, and how long the notice should display.

The update

Updates ship with editorial reasoning attached. Each affected artifact records:

• The triggering event (case name, date, citation).
• The prior version of the affected text.
• The current version.
• The reasoning for the update.
• The reviewer and review date.

For major shifts, a doctrinal notice surfaces at the top of every affected card for a defined period (typically thirty days), drawing reader attention to the change.

Why this matters

The Lab is not a snapshot. The doctrine the Lab interprets is alive and contested. A platform that did not update under triggering events would mislead readers about the current state of constitutional law. A platform that updated silently would erase the editorial record that makes its updates trustworthy. The protocol is the middle path: update fast, document the update, surface the change.

Every claim the Lab publishes is cited. The Lab uses a consistent citation framework across reform cards, bill records, Forge artifacts, doctrinal notes, and methodology content.

Manuscript anchors

Every Lab artifact includes a manuscriptCitations array that anchors the artifact to specific chapter, section, and line range in The First Three Words. This is the Lab's primary editorial accountability — the manuscript is the source of the structural argument the Lab operationalizes. A reader who wants to test whether the Lab's framing holds against the manuscript can follow the citation and check.

The manuscript citation format: Ch. [chapter] — [section name] (l. [line start]-[line end]). Example: Ch. VI — The Mechanism — The Sequence (l. 409-420).

External primary sources

For every external claim — a Supreme Court holding, a bill provision, a Federal Register entry, a state election commission record — the Lab cites the primary source with retrieval date and URL.

Verification status

Bill records and other artifacts that depend on time-sensitive primary-source data carry a verification status:

• VERIFIED. Dual-editor verification complete. The source URL has been fetched, the relevant claim confirmed against the primary text, and a second editor has independently checked. Records carry retrieval date.
• PENDING-PRIMARY-SOURCE-FETCH. The source URL is on file but verification is not complete. The artifact is published with the chip visible; readers can see what is and is not yet verified.
• DOCTRINAL CONCERN. A specific note flagging that the artifact's analysis depends on a doctrinal premise that has been contested or modified by recent ruling. The note is shown alongside the artifact.

Updating citations

When a primary source moves (a court opinion gets reported in an official volume, a bill version is amended, a CRS report is updated), the citation is updated and the change is logged in the artifact's edit history. The Lab does not silently update primary-source URLs; the change becomes part of the artifact's record.

What is not cited

The Lab does not cite anonymous sources, partisan press releases as primary authority, AI-generated summaries, or unsigned blog posts. These materials may inform editorial reading; they do not appear in citations.

The Lab is a particular kind of project. The clarity of its purpose depends on being explicit about what the Lab cannot do.

The Lab is not a litigation strategy. The Lab does not file cases, represent parties, or coordinate litigation. The Lab is the design surface for reforms; litigation is its own work, conducted by litigators.

The Lab does not pre-test reforms on real elections. The Engine and Simulator modules will model statutory and structural alternatives against documented data. They do not run controlled experiments on actual electoral outcomes. No platform can do that without changing the system it measures, which is the Lucas-critique-style failure mode the Lab is designed to avoid.

The Lab is at MVP preview. Multiple modules are in build (see /modules for current status). Where a module is not yet live, the Lab acknowledges it on the relevant surface — the Map module stub at /modules/map, the Forge "open soon" indicator, the bill cards' analytical panels labeled as launching in the coming phase.

AI analyses can be wrong. The Engine's reproducibility protocol is the first defense, the editorial review record is the second, the dispute process at /contribute#tier-classification is the third. The Lab does not claim AI-generated analysis is infallible. It claims AI-generated analysis is reproducible and reviewable.

The Lab is not neutral on the structural-failure thesis. The Lab operationalizes the argument The First Three Words makes — that the architecture of American institutions has stopped serving its citizens, that Article V is the founders' mechanism for redesigning architecture that has stopped serving the promise, and that the work of structural reform belongs to citizens. Readers who reject that thesis will reject the Lab's project. The Lab does not pretend to be neutral on this question. It does claim to be honest about the evidence it offers in support.

The Lab is not the deciding authority on what is constitutional. The Constitution is interpreted by the courts. Reforms become law through Article I, Article V, or state-level processes. The Lab models, drafts, and stress-tests proposals; the Lab does not decide.

The Lab cannot offer legal advice. Drafts, arguments, and analyses on the Lab are not legal advice and cannot substitute for professional legal counsel. Anyone considering filing a tier-dispute, drafting an amendment, or proposing legislation in their state or in Congress should consult qualified counsel.

Every artifact on the Lab carries a version. The Lab itself is at v0.1.0-alpha. Bill records, reform cards, Forge artifacts, and methodology content each have a draftVersion field visible in their metadata.

Version semantics

The Lab uses semantic versioning adapted for editorial artifacts:

• MAJOR (1.0+). A structural change to the artifact's core argument, classification, or scope. Triggered by tier reclassifications, doctrinal foreclosure, or rewriting around a different theory.
• MINOR (0.x). A substantive content update — new analytical panels, expanded reasoning, additional citations.
• PATCH (0.x.y). Corrections — typo fixes, citation updates, link repairs.

The change log

Every artifact's commit history is the change log. The full Lab change log is published at the GitHub repository.

Doctrinal version increments

When a doctrinal-change protocol review (see above) results in an update, the affected artifact's MAJOR version increments — even if the textual change is small. The increment is the signal that the artifact's editorial premise has been re-examined under new authority.

Reading versions

Readers can see an artifact's current version in the metadata footer of any reform card, bill record, or Forge artifact. Comparing versions across an artifact's history is available via the GitHub repository.

Argue from primary sources. Claims that cannot be supported from the primary record are not published. Editorial judgment is welcome and explicitly marked when present; it is not dressed up as fact.

Acknowledge uncertainty. Where doctrine is contested, where data is incomplete, where reasonable people read the evidence differently, the Lab says so. Apparent confidence is reserved for places the evidence supports it.

Mark editorial judgment as such. The Lab's tier classifications, the Lewis Act post-Callais reading, the Engine's reasoning chains — all involve editorial judgment. When judgment is the operating mode, the Lab labels it. Readers do not have to guess.

Update when wrong. The Lab maintains a public version history. When the Lab's analysis is wrong, the correction is documented, the prior version remains accessible in the commit history, and the editorial reasoning for the change is published.

Do not perform false neutrality. The Lab is not neutral on whether the structure has failed (it has) or whether Article V is a legitimate path (it is). The Lab is rigorously non-partisan — it does not endorse parties, candidates, or political coalitions — but it is not neutral on the project of constitutional repair.

Make the editorial process visible. Every editorial decision is reviewable. Reviewers are named (or, for pseudonymous panel members, named by role). The reasoning for each decision is published. The process for disputing a decision is documented at /contribute#tier-classification.