# Responsible America — Research Methodology Status: public-facing. This document explains, in plain terms, how Responsible America selects topics, ranks sources, verifies facts, handles disagreement, evaluates reforms, treats numbers, corrects itself, and uses automation. It is written to be read by anyone — supporters, critics, and journalists alike — who wants to know how a page on this site was built. Last reviewed: 2026-07-30. ## 1. How topics are selected Topics are not added arbitrarily or because a subject is trending. Every reform on the site maps to a specific place in a fixed taxonomy: - **17 divisions** (`div-01` through `div-17`) group the platform into broad areas — for example Division 4, "Corporate Power and Political Money," or Division 10, "Healthcare and Prescription-Drug Reform." - **20 sections** (lettered A–T) sit under those divisions and define the actual subject areas a reform can belong to — elections, campaign finance, lobbying, corporate power, tax, executive power, Congress, courts, mental health, education, healthcare, banking, work, housing, environment, infrastructure, AI and strategic competition, military and foreign policy, secrecy/intelligence/UAPs, and debt/government management. Each section carries a defined list of subsections (for example, section K, "Healthcare and Prescription Drugs," includes subsections such as `prior-authorization`, `pharmacy-benefit-managers`, and `rural-healthcare`). - Every reform record's `section` field must reference one of these section ids, and its optional `subsection` field, when set, must reference a slug already listed under that section. A topic that does not fit an existing section is not force-fit into one — the taxonomy is treated as extensible (`data/policy-sections.json` notes it is "not a permanent ceiling"), and a new section or subsection is proposed and added deliberately, not invented ad hoc inside a single reform record. Subjects that are not yet developed enough to sit in the numbered American Renewal Plan (Level 2) go into the **Issues Laboratory** (Level 3, `div-16`). The Issues Laboratory has its own seed list of undeveloped topics — immigration, policing and criminal justice, gun policy, abortion and reproductive policy, Social Security and Medicare solvency, agriculture, tribal sovereignty, cryptocurrency and digital money, and similar subjects where research, public discussion, legal review, or fiscal analysis is still in progress. A Level 3 topic never graduates to Level 2 "because it sounds attractive." It must clear the same legal, fiscal, and editorial review described in `docs/editorial-policy.md` with real, non-placeholder answers to the Responsible America Test (Section 6, below) before it is treated as part of the platform. ## 2. How sources are ranked Responsible America uses a fixed source hierarchy, from most to least authoritative. The full 15-tier list, with the rules for using it, is maintained in `docs/source-policy.md`; the ordering, in brief, is: 1. The Constitution and official constitutional materials 2. Statutes and regulations 3. Court opinions 4. Official government datasets 5. GAO reports 6. CBO analysis 7. CRS reports 8. Inspector General reports 9. Federal and state agencies 10. Legislative records 11. Peer-reviewed research 12. Established nonpartisan research institutions 13. Reputable journalism 14. Advocacy organizations representing competing positions 15. Public commentary and social media, used only as evidence of public reaction, never as evidence of fact A claim about what a statute says is checked against the statute itself, not a news article summarizing it, whenever the statute is available. Journalism and advocacy material are used to establish context, framing, and the existence of a dispute — not to establish disputed facts that an official source could settle directly. ## 3. How facts are verified Every material factual claim used in a reform is captured as its own claim record (see `docs/fact-checking-policy.md` for the full field list), separate from the prose that cites it. Each claim record carries a `verification_status` of one of: `verified`, `verified_with_qualification`, `disputed`, `outdated`, `pending_review`, `unsupported`, or `rejected`. A claim moves between these states only on evidence, not on convenience: - **`pending_review` → `verified`**: at least one source at or above tier 9 of the hierarchy (agency, legislative record, or higher) directly supports the claim as stated, with no credible contradicting source found during a documented search. - **`verified` → `verified_with_qualification`**: the claim is substantively correct but requires a caveat — a partial figure, a jurisdiction-specific result, or a methodology limitation the source itself flags. - **`pending_review` or `verified` → `disputed`**: a credible source (tier 1–13) is found that directly contradicts the claim, and neither side's evidence is clearly superior. The `contradicting_source_ids` field is populated and both interpretations are represented in the claim's `public_notes` where relevant. - **`verified` → `outdated`**: a newer authoritative figure supersedes the one on record. The old claim is not deleted (see Section 8). - **`pending_review` → `unsupported`**: after an actual documented search, no source at any tier could be found that supports the claim as written. It is not published as fact. - **Any state → `rejected`**: an active source directly and credibly refutes the claim, or the claim is found to be fabricated, misattributed, or a misreading of its cited source. Claims are never marked `verified` because a claim "sounds true" or matches the platform's thesis. The evidentiary bar is the same regardless of which political direction the claim cuts. ## 4. How conflicting evidence is handled When serious, credible sources genuinely disagree on a controversial claim — a disputed cost estimate, a contested legal interpretation, a disputed effect of a past policy — Responsible America does not silently pick a side. The claim is marked `disputed`, both `source_ids` and `contradicting_source_ids` are populated, and the public-facing text presents the credible sources behind each major competing interpretation rather than asserting one as settled. This is different from false balance: it applies only where the underlying sources actually conflict, not where one side has evidence and the other has only assertion (see Section 5). ## 5. How political balance is approached Balance does not mean treating all claims as equally supported. It means applying the same evidentiary standard to every claim regardless of which side it helps or hurts. If a well-evidenced claim is politically inconvenient to one side, it is presented as well-evidenced — the methodology states this openly rather than hedging a claim that has actually cleared verification. Conversely, a poorly-evidenced claim is not published as fact merely because it is popular or convenient to either side. "Balance" on this platform refers to giving competing *interpretations* a fair hearing when the underlying evidence is genuinely mixed (Section 4) — not to manufacturing symmetry where none exists in the evidence itself. ## 6. How reforms are evaluated — the Responsible America Test Every reform, before it can be treated as a serious platform position, must have real answers (not placeholders) to all 28 questions of the Responsible America Test: 1. What exact problem is being solved? 2. How large is the problem? 3. What evidence demonstrates it? 4. Who benefits from the present system? 5. Who bears its costs? 6. What would the reform change? 7. What would it leave unchanged? 8. What constitutional authority supports it? 9. Does it require an amendment? 10. Can Congress implement part of it by statute? 11. Can states act independently? 12. What would it cost? 13. How would it be funded? 14. Who would benefit? 15. Who could be harmed? 16. Could it reduce economic growth or innovation? 17. Could it be abused by a hostile administration? 18. Does it protect due process? 19. Does it increase or reduce concentrated power? 20. What civil-liberties concerns exist? 21. What is the strongest supporting argument? 22. What is the strongest opposing argument? 23. Is there a workable compromise? 24. What unintended consequences are foreseeable? 25. How will success be measured? 26. When will it be reviewed? 27. Should it expire automatically? 28. Would it remain fair under opposing-party control? Question 28 is treated as a hard filter, not a formality: a reform that would only be tolerable while one party or the other holds power fails this test and is not published as an approved position. ## 7. How fiscal claims are treated Fiscal figures on this platform are ranges built on stated assumptions, not false precision. Each fiscal estimate on a reform record (`estimated_federal_cost`, `estimated_state_cost`, `estimated_private_cost`, `estimated_savings`) carries a `status` of `estimated`, `range`, `pending_review`, or `not_applicable`, plus a `low`/`high`/`point` figure, a `unit`, a `period` (e.g., "annual," "over 10 years," "one-time"), and a `basis` describing the methodology or source behind the number. Two distinctions are enforced on every fiscal claim: - **Estimated vs. guaranteed savings.** A savings figure is a modeled projection under stated assumptions, not a promise. Public text uses language like "estimated to save" or "projected range," never "will save" or "guarantees," unless a savings figure comes from realized, reported outcomes rather than a forward projection. - **Policy modeling vs. official government scoring.** An internal Responsible America cost estimate is labeled as Responsible America's own modeling. A CBO, GAO, or Joint Committee on Taxation score is labeled as such and is not blended with or presented as interchangeable with internal modeling — the two are cited and captioned separately. ## 8. How corrections are made Every reform record carries a `change_log` — an append-only array of dated entries, each with a version number, a summary of what changed, and (where applicable) a reason and reviewer. Corrections are made by adding to this log, not by quietly rewriting a page. When a figure becomes outdated, the old data point is not deleted. It is marked `outdated` in its claim record and preserved with the `as_of_date` it was accurate for, while the public-facing text is updated to the latest verified figure. This means the site's current text always shows the latest number, but the record of what was previously stated — and when it changed — remains queryable. ## 9. How automation is used, and its limits Automated and AI-assisted research is used to gather public source material, draft initial claim and reform records, and flag contradictions or gaps for a human reviewer. Automation produces a review queue — it does not publish conclusions to the public site. Every record an automated process touches is created or left in `publication_status: draft` (or `internal_review`) with `fact_check_status: pending_review` until a qualified human reviews it. Full rules for AI-assisted work on this project are in `docs/ai-usage-policy.md`. ## 10. How human approval works Every reform record moves through a fixed editorial workflow before it is public: `draft → research_review → legal_review → fiscal_review → editorial_review → approved → published → archived` The requirements to move between each of these states, and who is authorized to move a record, are defined in full in `docs/editorial-policy.md`. In brief: a record cannot reach `published` until it has passed legal review (constitutional/defamation/framing issues), fiscal review (cost and savings figures checked and sourced), and editorial review (structure, language rules, and balance checked), in that order, with a named reviewer recorded at each stage. ## 11. How citizens may report errors Readers who believe a page contains an error, an outdated figure, or a missing perspective can flag it through the correction-request mechanism described in `docs/citizen-action-policy.md`. That document is the authoritative reference for how correction requests are submitted, routed, and resolved; this document does not duplicate it.