What is "governance"?
Three words get used interchangeably, but they mean different things.
Politics sets the goals. Government does the work. Governance is the set of rules that shapes how that work gets done: how a bill reaches the floor, how votes are cast and recorded, how a law is written, how an agency carries it out, and how anyone finds out whether it worked. Politics decides what we want. Governance determines what is actually produced.
This project stays out of politics on purpose. Reasonable people disagree about goals, and in a democracy those disagreements are settled by voting. What we study is the machinery, and whether it turns decisions into results no matter who wins. The same machinery serves every party that holds power, and when it is badly designed, it fails all of them.
The line is not perfectly clean. Some design choices carry political weight of their own. When they do, the aim is to state them openly, so they can be argued about on their merits, rather than bury them in the drafting.
Why this is needed
Aviation, medicine and automobile safety all improved the same way. When something failed, people investigated, traced the failure to a cause, changed the design, and checked whether the change worked. Over decades, that loop made each of them dramatically safer.
Government mostly lacks that loop, at both ends. Before a law passes, it is rarely designed the way engineered systems are: the incentives it creates, and whether its mechanism actually serves its goal, are seldom worked out in advance. After it passes, there is usually nothing to check it against. When a law goes wrong, the argument that follows is about who is to blame, not about what in the design failed.
Governance science has two main parts: design, getting the rules right before they take effect, and measurement, checking afterward whether they did what they were meant to do.
Design: getting the rules right
Rules shape behavior. People respond to the incentives a rule creates, whoever they are. A well-designed rule points those incentives toward its goal. A poorly designed one rewards something else, and capable, well-meaning people working inside it still produce bad results.
Design work asks a few questions before a rule takes effect:
- What is it for? (the mandate)
- Does the mechanism actually serve that goal? (the architecture)
- What will the people it touches be rewarded for doing?
- What else will it do, and what will it cost?
- How will anyone know whether it worked?
The bootstrap problem
There is a catch. The laws that apply to citizens and businesses are written by a government that runs on rules of its own: how bills reach the floor, how committees work, how votes are recorded, how agencies are overseen. Those internal rules shape every law that comes out, so they need the same design discipline as the laws themselves. The work has to start with the rules that shape the rules, which is why much of this project's research focuses on how Congress itself operates.
To help with all of this, the project is building a library of design patterns: recurring design problems, and approaches that have worked for them elsewhere.
Measurement: checking whether it worked
A law usually spells out what to do, but it rarely says what doing it should accomplish, in terms anyone could check later. Without a stated goal, there is nothing to measure against.
Measurement means stating the goal up front, deciding in advance what evidence would show success or failure, and then looking. When a law falls short, it also means locating the failure: was the goal wrong, was the design wrong, or was it carried out badly? Each has a different fix, and without a stated goal, they are impossible to tell apart.
Utah's 2025 "100% ID law" is a recent example. Its sponsor later said the way it operated "was never the intent of the bill," but the intent was never written into the law, so no one could tell whether it worked. Read the 60-second version →
The method
The approach is the one every applied science uses, adapted to laws and institutions:
Observe
Start from outcomes people can see: shutdowns, laws that miss their target, reforms that backfire.
Separate the parts
Every law bundles three things: the mandate (what it is for), the architecture (how it is supposed to achieve that), and the execution (whether it was carried out as written). Most failures hide in the gaps between them.
Trace the mechanism
Identify which rule or incentive produces the outcome, regardless of who holds office.
State what would prove it wrong
Every claim comes with the evidence that would refute it, written down before the evidence is examined.
Test and measure
Check the claim against real cases, with methods other people can repeat.
Keep the record
Publish results either way, and correct mistakes openly rather than quietly revising.
None of this removes values from government. Deciding what a law should accomplish is a political question, and it should be. The claim is narrower: once the goal is stated, whether a design achieves it can be checked.
How the work holds itself to account
A project that faults government for never checking its own work has to check its own. If we ask laws to state their goals and be tested against them, our claims have to meet the same standard. Otherwise this is just one more opinion about government.
It also means the work can be judged without trusting us. Anyone can see how strong the evidence is behind a claim, and what would prove it wrong. Every significant claim carries a label saying how strong the evidence is:
- Claims state what would prove them wrong.
- Test rules are fixed before testing, and completed research records are frozen. Corrections go in a separate, dated erratum.
- The research is public and free to reuse under a Creative Commons license (CC BY-SA 4.0).
- It is nonpartisan by design. The work diagnoses rules and incentives, not people or parties.
The Governance Design Agency
The project's flagship proposal puts both parts, design and measurement, into one institution. The Governance Design Agency (GDA) is a proposed institution with a single function: designing the how, the implementation that pursues a goal elected officials have set.
You tell us what you want to achieve. We will tell you what would achieve it, what it would cost, and what else it would do. Then you decide.
How it would work:
Politics sets the goal
Elected officials decide what society wants to achieve, stated so that success can be checked.
The GDA designs the options
It produces designs that would achieve the goal, with their costs, benefits and side effects spelled out.
Politics chooses
Elected officials pick a design, change it, or reject them all, and pass the result into law. The executive branch carries it out. GDA designs carry legal force only where a legislature adopts them.
The GDA keeps the record
It publishes what it recommended, what was decided, what was carried out and the reasons given, and provides that record to voters. When something fails, they can trace the failure to the goal, the design, or how it was carried out.
The GDA does not set goals, implement, enforce or judge anything. It is a long-term proposal, not a near-term one. The current specification is published, along with a candidate mandate for the agency itself, written so that whether the GDA achieves its purpose could be checked, and offered for anyone to argue with.
Read the specification (v4.0, for paid subscribers; summary free) →
What exists so far
The work has four layers. Some are built and tested; some are early. The status is stated plainly.
| Layer | What it is | Status |
|---|---|---|
| Notation | The mandate / architecture / execution distinction: a way to read any law and see which part is which. | Written, being formalized |
| Method | A written protocol for analyzing a single bill or executive order from its primary text, used for Church Bells briefs. | In use |
| Measurement | A scoring system for how much government machinery an executive order deploys, so orders can be compared. Research repository. | Validated |
| Catalogue | A shared vocabulary and a library of design patterns: recurring problems and approaches that have worked elsewhere. Research repository. | Early: 11 terms, 5 patterns |
Research
Current and planned studies. Each is labeled by how far along it is.
How much machinery does an executive order deploy?
ValidatedA scoring system that reads an executive order's text and measures how much government machinery it puts in motion, so orders can be compared on something other than headlines. The scoring rules were written down and fixed before testing. They were applied to 298 executive orders, which were then re-scored from scratch in a separate blind pass, without access to which ones the political science literature classifies as significant (Mayer & Price, 2002). The blind scores matched that classification with an AUC of 0.78, where 0.5 would be a coin flip and 1.0 would be perfect. Next: extending the scoring to the full corpus of executive orders.
The captured wave: do new members change Congress?
UnderwayIf sending new people to Congress reduced partisanship, elections with more turnover should be followed by less party-line voting. Comparing House reelection rates with party-line voting from 1970 to 2024 suggests the relationship has changed. In the 1970s and 1980s, the two were only loosely connected. Since the mid-1990s, they have moved in opposite directions: when more incumbents lose, party-line voting goes up, not down. If that holds, the institution now shapes new members faster than they can change it, which bears directly on proposals like term limits. Preliminary, using Brookings Vital Statistics on Congress data.
The coercion loop: recorded votes and party discipline
Paper in circulationThe 1970s reforms put individual members' votes on the House floor, including votes on amendments, on the public record, for good reasons. Combined with party leadership's control over members' careers, the same transparency made every member's position visible to leadership in real time. The paper traces how that combination rewards compliance with leadership over independent judgment, regardless of who is elected.
Who checks whether laws worked?
UnderwaySeveral democracies already have pieces of the loop. Switzerland's constitution requires Parliament to ensure federal measures are evaluated for effectiveness. The United Kingdom's government commits to reviewing Acts three to five years after they pass. Sweden sets objectives for its agencies in annual instruction letters, and its national audit office follows up on its recommendations. Estonia has changed how its laws are drafted, and New Zealand's leaky-buildings crisis showed what happens when a performance-based building code has no one checking the results.
The early finding is consistent across these cases: the capability to evaluate exists, and evaluations get produced, but they rarely reach the one party that can act on them. That party is the constituents. Legislators and agencies can set an unwelcome evaluation aside. Voters can respond to one only if they can see it and trace the problem to its source. This is why the GDA records the designs, what was decided and what was carried out, and provides that record to voters. With it, they can tell whether a failure came from the goal, the design or its execution, and exercise their democratic accountability if they choose to.
The price of saying no
Early draftA theoretical foundation for when a rule is genuinely consented to: how much a participant is penalized for withholding agreement, and why that price, more than who holds office, determines how freely people can dissent inside an institution.
Organizational research already measures something close to this inside companies. Muel Kaptein's Corporate Ethical Virtues model (2008) assesses, among other dimensions, whether people can raise concerns openly ("discussability") and whether wrongdoing has consequences ("sanctionability"). The working hypothesis is that the same dimensions can be applied to government institutions, to the designs the GDA would produce, and to the GDA itself.
Read the work
- The Rules that Shape the RulesA book, published one chapter at a time, that walks through finding the source of Congress's dysfunction from observable evidence.
- A Law That Nobody Could CheckThe Utah case in full, and the institution it points to.
- The Statecraft BlueprintEssays and papers applying the method to specific parts of government.
- EO Structural WeightThe executive-order scoring system: the pre-registered rules, the 298-order validation, and every score with its justification.
- Governance ScienceThe shared vocabulary, design patterns and foundations, and the source of this website.
- Church BellsBriefs on new bills and executive orders, read from the primary text.
Work with us
Governance science is the method. The Statecraft Blueprint is the business that applies it: we design governance mechanisms, both legislative text (defining mandate and architecture) and the rules a government operates on. Whether anyone uses a design is up to them, and only a government can provide the execution.
- Have a government failure you’d like to see fixed? Contact us to discuss a design.
- Work in government? Answer one question: why can’t our government improve?
- Subscribe to The Statecraft Blueprint to read new work as it’s published. Paid subscribers get the full specifications.
- Know someone who should see this? Share this page.