How this project will handle evidence and disagreement

A policy publication can mislead without inventing a number. It can select only favorable evidence, use a national average to erase local variation, present an advocacy brief as settled research, turn correlation into causation, or describe a proposal in the grammar of existing law. It can hide uncertainty behind confident design.

The Next Square Deal is built to make those habits harder.

Sources have different jobs

A primary historical document can tell us what an institution or speaker said. It cannot by itself prove that a policy worked. A statute can tell us what Congress enacted. It may not describe current implementation after amendments, regulations, cases, and administrative practice. A government dataset can support a measure while leaving causal questions open. A government audit can reveal implementation failures without establishing that every alternative would perform better.

The evidence library therefore labels both source type and source role. A source may be supporting, critical, contextual, historical, or methodological. Those labels are not grades. They tell the reader why the item is here.

For example, HUD’s worst-case-housing-needs report uses a defined measure for a defined renter population. It is strong evidence about that measure and population, not permission to turn the result into every claim about housing. NCES education indicators describe conditions and trends, but descriptive data do not by themselves identify a policy’s causal effect. CMS expenditure accounts map where health spending flows, not whether every patient can obtain care. U.S. Department of Housing and Urban Development, Office of Policy Development and Research 1 National Center for Education Statistics 2 Centers for Medicare & Medicaid Services 3

“Government source” does not mean “beyond criticism”

Official sources can provide unique authority, methods, administrative data, and primary text. They can also reflect statutory definitions, agency scope, measurement choices, lag, and institutional incentives. This project will describe material limitations and use independent or critical evidence where it adds a real test.

Government audits are especially useful because they examine whether institutions perform as promised. GAO’s review of unemployment insurance addressed program design, state administration, technology, access, timeliness, and integrity. Its FOIA review examined selected agencies’ compliance with proactive-disclosure requirements. Neither report proves that government always fails. Both show why capacity and implementation belong in policy design from the beginning. U.S. Government Accountability Office 1 U.S. Government Accountability Office 2

Evidence is not the same as judgment

Evidence can estimate effects, describe institutions, identify distribution, or test mechanisms. It cannot decide every question of value. Two readers may accept the same estimate and disagree about whether a cost is justified, which level of government should act, or how much uncertainty is tolerable.

The publication will mark that boundary. “The source reports” is different from “this project recommends.” “Evidence suggests” is different from “the policy guarantees.” A projection is different from an observed result. A voluntary framework, such as NIST’s AI risk-management framework, can offer a useful method without creating a legal duty. National Institute of Standards and Technology

Serious objections belong on the page

Each guarantee includes at least two objections stated in a form a thoughtful critic could recognize. Concerns about fiscal cost, administrative capacity, market response, federalism, privacy, fraud, work incentives, supply, transition, or regulatory capture are not evidence of bad character. They are design tests.

A steelmanned objection may lead to a narrower proposal, a different institution, a pilot, a safeguard, a compensation mechanism, or a decision not to proceed. It may also be outweighed by other evidence and values. The publication will explain that judgment rather than pretending the objection disappeared.

Working draft means visible uncertainty

A working-draft label does not excuse an unresolved interface, a fabricated citation, or a fake statistic. It means the framework remains open to research and revision. Where a legal survey is incomplete, the page should say so. Where current data have not been reviewed, it should avoid the number. Where evidence is mixed, it should describe the disagreement and method.

Draft sources may be cataloged for future verification, but they cannot support a published factual claim. Citation IDs are validated so a renamed or missing source fails the content check rather than becoming a broken reference.

Corrections and revision

A factual correction is not an embarrassment to bury. Material changes should be dated and explained. Review dates should be visible. Archived links should be added where useful. Legal-status labels should change when law changes. A policy recommendation should change when stronger evidence or implementation experience changes the balance.

The goal is not to build a publication that never admits error. It is to build one whose structure makes error easier to find, correct, and learn from.