# Responsible America — Fact-Checking Policy

Status: internal operating procedure for anyone verifying, disputing,
or updating a factual claim on this platform.

Last reviewed: 2026-07-30.

## 1. The claim record model

Every material factual claim used anywhere on the site is captured as
its own record, separate from the reform prose that cites it, per
`data/schemas/claim.schema.json`:

| Field | Purpose |
|---|---|
| `claim_id` | Stable id, pattern `claim-[a-z0-9-]+`. |
| `claim_text` | The exact claim being made, in a form that can be checked true/false against a source. |
| `claim_type` | One of `statistic`, `legal_holding`, `legislative_status`, `fiscal_figure`, `historical_fact`, `quotation`, `poll_result`, `other`. |
| `related_reform_ids` | Which reform record(s) this claim supports (minimum one). |
| `source_ids` | The source(s) that support the claim (minimum one). |
| `support_level` | `directly_supported`, `reasonably_inferred`, `contested_interpretation`, or null. |
| `contradicting_source_ids` | Any sources found that conflict with the claim. |
| `date_range` | The period the claim covers, if applicable (e.g., "FY2019–FY2023"). |
| `jurisdiction` | The jurisdiction the claim applies to (federal, a specific state, etc.), if applicable. |
| `as_of_date` | The date the underlying figure/fact was accurate as of. |
| `last_verified` | The date a reviewer last checked this claim against its sources. |
| `verification_status` | See Section 2. |
| `reviewer` | Who last verified or changed the status. |
| `public_notes` | Reader-facing explanation of uncertainty, qualification, or dispute. |
| `internal_notes` | Reviewer-facing notes only — never rendered publicly (Section 5). |

A claim record must have at least one `related_reform_ids` entry and at
least one `source_ids` entry before it can leave `pending_review`. A
claim with no source is not a claim record — it is a note for a
researcher to go find one.

## 2. Verification statuses and their evidentiary bar

| Status | Bar |
|---|---|
| `verified` | At least one source at tier 9 or higher (agency, legislative record, peer-reviewed research, GAO/CBO/CRS, IG, court opinion, statute/regulation, or the Constitution — see `docs/source-policy.md`) directly and unambiguously supports the claim as written, and a documented search turned up no credible contradiction. |
| `verified_with_qualification` | The claim is substantively correct but needs a stated caveat — a partial figure, a specific jurisdiction, a methodology limit the source itself notes, or a scope narrower than the claim's plain reading would suggest. |
| `disputed` | A credible source (any tier) directly contradicts the claim, and the conflict is genuine — not one side simply being wrong. `contradicting_source_ids` is populated and both interpretations are represented in `public_notes` where the page discusses the claim. |
| `outdated` | The claim was previously verified but a newer authoritative figure now supersedes it. The claim record is not deleted; see Section 3. |
| `pending_review` | Default status for a newly drafted claim. No reviewer has yet checked it against sources. |
| `unsupported` | A documented search was performed and no source at any tier could be found supporting the claim as written. Not published as fact. |
| `rejected` | An authoritative source directly refutes the claim, or the claim is discovered to be fabricated, misattributed, or a misreading of the source it cites. |

A reviewer changing `verification_status` sets `reviewer` and
`last_verified` in the same edit. A status change without a named
reviewer is not a valid edit.

## 3. Outdated data is preserved, never deleted

When a newer authoritative figure supersedes a previously verified
claim, the original claim record's `verification_status` is set to
`outdated`, its `as_of_date` is preserved exactly as it was (so it
remains clear what period the old number described), and a new claim
record is created for the current figure. The reform page's visible
text is updated to cite the new claim record's current figure. The old
record stays in the data store and remains queryable — a reader or
reviewer can always see what was previously stated, when it was
accurate, and when it changed. This mirrors the `superseded_by`
mechanism used for source records (`docs/source-policy.md`).

## 4. Fact-check checklist — run before marking a claim `verified`

A reviewer works through each of the following distinctions explicitly
before advancing a claim past `pending_review`:

- **Allegation vs. adjudicated finding.** Is this something a court,
  agency, or Inspector General has actually found to be true, or is it
  an accusation still being investigated or litigated? If the latter,
  the claim text says "alleged" and cites the accusing source, and
  `verification_status` reflects that it is not an adjudicated fact.
- **Proposed bill vs. enacted law.** Does `claim_type: legislative_status`
  describe something that has passed both chambers and been signed (or
  enacted over veto), or is it still pending? A bill's current stage is
  stated precisely (introduced, passed committee, passed one chamber,
  passed both chambers, signed, enacted) using the same categories as
  `current_legislation.status` on the reform record.
- **Trial-court vs. binding appellate ruling.** Is the cited holding
  from a trial court (persuasive only, can be reversed) or from an
  appellate court whose ruling binds the relevant jurisdiction? A
  trial-court holding is not described as settled law.
- **Campaign promise vs. completed action.** Did an official actually
  do the thing, or only say they would? A promise or statement of
  intent is captured as a `quotation` claim type, not as a
  `historical_fact` about an action taken.
- **Announced spending vs. actual obligations/outlays.** Is the figure
  a budget request, an authorized/appropriated amount, an obligated
  amount, or money actually spent (outlays)? These are different
  numbers and are not used interchangeably; the claim text specifies
  which one it is describing, sourced to the government dataset that
  actually reports that stage.

A claim that fails any of these distinctions is not marked `verified`
until the claim text is rewritten to match what the evidence actually
supports.

## 5. Public vs. internal notes

`public_notes` is reader-facing: it explains uncertainty, qualification,
or the nature of a dispute in language a reader can use to understand
why a claim is marked the way it is. `internal_notes` is
reviewer-facing only — reviewer disagreements, leads not yet followed
up, concerns about a source's reliability that haven't been confirmed,
or process notes. The site layer must never render `internal_notes`
publicly; this mirrors the same rule for `editorial_notes` on the
reform schema. Anyone building a page template that surfaces claim data
checks this before wiring a new field into the public site.

## 6. Named-person claims (Phase 4)

A claim about a specific, identifiable person (a sitting or former
official, a judge, a justice) carries additional required fields:
`claim_nature` (`accusation` / `description_of_official_finding` /
`description_of_court_holding` / `policy_criticism` /
`factual_background`), `official_capacity`, `official_finding`,
`adjudicated`, `alleged`, `disputed`, `response_available`,
`legal_review_required`, and `public_interest_basis`. The rules:

- `official_finding: true` is set **only** when an official body (a
  court, an ethics committee, an inspector general) actually made that
  finding — never for an allegation alone, no matter how well-reported.
- A claim describing an allegation, controversy, or reported concern
  that has not been adjudicated is `alleged: true`, `adjudicated: false`,
  and its `claim_nature` is `accusation` or `policy_criticism` as
  appropriate — never `description_of_official_finding`.
- Public rendering (see `js/components.js`'s `namedPersonStatusLine`)
  preserves attribution and status — "alleged, not adjudicated,"
  "disputed," "official finding" — rather than either hiding the
  distinction or overstating it.
- `named_person_ids` is an internal tracking array only (not a public
  profile — none is created in this phase) and is stripped before any
  public template ever sees it; see `js/site.js`'s
  `stripClaimInternalFields`.
- `legal_review_required: true` is set on every named-person claim that
  touches potential wrongdoing, triggering the publication gate
  described in `docs/legal-review-policy.md` section 6.
- No claim states that a named person committed a crime, is corrupt, or
  violated ethics rules as settled fact unless an authoritative
  adjudication or official finding actually supports it.

## 7. Health and science evidence (Phase 5)

A claim citing a study or survey (`data/schemas/claim.schema.json`'s
`study_design` field and its neighbors — `population`, `sample_size`,
`effect_measure`, `effect_size`, `confidence_interval`,
`replication_status`, `causal_or_associational`, `peer_reviewed`,
`funding_source`, `conflict_of_interest`) must set `study_design`
accurately (a government survey is `government_surveillance_data`, not
a randomized trial; an association found by comparing groups after the
fact is `observational_cohort` or `case_control`, not causal by
default) and must set `causal_or_associational` honestly whenever
`study_design` is set. **An observational association is never
presented as proof of causation.** This is the single most common way
health and social-science claims get overstated, and it is treated as
a fact-checking failure exactly like a fabricated statistic — not a
minor wording issue.

These fields are not required for a non-study source (a statute, a
government dataset citation, a court opinion) — leave them null.

See `docs/security-and-privacy.md` section 10 for the related privacy
rules governing health, family, and child data specifically.