How Fonteum decides what to publish.

Fonteum publishes research and data drawn entirely from federal — and a handful of state — public-records sources. Editorial decisions — what to study, which datasets to ingest, how a federal record is displayed, when corrections are issued — are made independently of who pays Fonteum for data access.

Paid API access and paid pilots buy production rate limits, support, and custom dataset delivery — nothing else. They do not buy a research finding, do not change how a federal record is displayed, and do not delay a correction. Fonteum does not rank, rate, or place providers, so there is no placement to sell.

• No paid placement — Fonteum does not rank or rate providers
• No advertiser veto on study scope or findings
• Paid pilots and API tiers buy data access only, never editorial outcomes
• Corrections are never delayed for a paying customer

Provider data is sourced from federal public records (CMS, HHS-OIG, HRSA, AHRQ, BLS, BEA, Census) and a handful of state licensing and Medicaid-exclusion feeds. The public data-provenance register documents source mappings; record-level source fields are identified where available and can be null or absent.

Research figures are sourced from those same federal and state public datasets and study-specific primary sources cited inline in the methodology block of each study. We do not cite commercial aggregator sites as primary sources — they aren't.

Named authorship is policy. Every published study carries a byline that links to the author's profile. The author is responsible for the figures in the piece; the editor reviewing it is named in the masthead. Anonymous bylines are reserved for short site-operations notes (corrections-log entries, factsheet refreshes) where individual attribution would be theatre.

Fonteum Research is the institutional byline for studies with multiple contributors. The editor of record for any institutionally-bylined study is identified in the methodology block of that study.

Editors and authors disclose any financial interest in a provider, facility, or category before they cover it. The standing disclosure: Fonteum's revenue comes from paid API access and paid data pilots; no customer relationship influences what we study or how a federal record is displayed. The standing disclosure is restated in the methodology block of every study.

Per-piece disclosures (a contributor holds equity in a healthcare company, a contributor's spouse works for a covered facility, a contributor previously consulted for a data customer) are stated at the top of the affected piece.

Publication policy requires (1) a numbers check — the editor-of-record reproduces the headline figures from the dataset, (2) a sourcing check — the editor records the source's available loaded or observation date and does not treat publisher cadence as proof of currency, and (3) a tone check against this policy and the methodology page.

After publish, the study sits in a 7-day correction window during which any reader-flagged inaccuracy is treated as a P1 issue. Corrections accepted during the window are noted on the study page and entered into the corrections log.

When a published figure is wrong, we correct it. The original page is updated, the changed sentence carries a `[corrected on YYYY-MM-DD]` marker, and the substance of the correction is logged in the public corrections log. We do not silently re-publish a piece with revised numbers.

When current production data cannot support a study's derivation, we remove it from active discovery and mark it pending withdrawal while an operator reviews the evidence. A pending withdrawal is not a final retraction; the historical page remains labelled at its original URL during the decision.

Reader-submitted corrections route through the public corrections form on this site and on every vertical-domain. We respond to every submission. We name the cause of each accepted correction (data refresh, source change, editor error, owner-claim update) so readers can weight the failure mode.

If a study's premise — not just a figure — turns out to be wrong, the study is retracted. The retraction notice replaces the study text, names the editor of record, names the substantive failure, and remains at the original URL so external citations don't 404.

Finalized retractions are listed in the corrections log alongside corrections, with their own status flag. Pending withdrawals carry a separate status until an operator decides. We have not finalized a study retraction to date; the policy exists so the threshold is documented, not tested.

Authors may use AI assistants to draft prose, generate exploratory queries, or scaffold research code. AI does not approve a study for publish; a named editor does. AI-generated text is reviewed line-by-line against the source data; any number that survives review traces back to the dataset, not to a model's prior.

We do not ship AI-generated images of people, AI-fabricated quotes, or AI-summarised business reviews. Visualisations on study pages are produced from the dataset by a named author.

The word "verified" — and its adjective family — is forbidden in user-facing copy on every Fonteum page. Figures are checked. Fonteum does not "verify" a provider — it republishes what a named federal or state source publishes. Policy requires the available source citation and snapshot or observation date to be shown; when the record omits one, the page must disclose the missing value rather than invent it.

We tier sources by who publishes them, not by who cites them. Tier-2 (profile enrichment) sources are federal or state agencies with a published distribution license that permits attribution-required republication of the fields we surface — currently CMS NPPES, CMS PECOS, CMS Care Compare, OIG LEIE. Tier-1 (research-only) sources are federal datasets we cite for context but do not republish per-row — currently BLS OEWS, BEA personal income, U.S. Census, HRSA HPSA. Board-certification and credentialing facts are named in coverage only when a federal record itself carries them; third-party credential registries are not republished.

Sources outside this taxonomy do not enter the graph. Commercial aggregator sites, broker lists, lead-resale databases, and screen-scrape outputs of restricted registries are explicitly excluded — see the §125 RESTRICTED_SOURCES list at /sources for the standing exclusions (state medical-board registries, NMLS Consumer Access, ABMS / CertiFacts, Google Places backfill).

• Tier-2 (profile enrichment): federal/state, attribution-required
• Tier-1 (research-only): federal context, not republished per-row
• Credentialing facts: cited only from federal records, never re-registered
• Restricted-source list at /sources is the explicit exclusion register

Disagreements happen — a state licensing board lists an address that doesn't match the address the same provider's NPPES record carries. The default rule: when a primary-issuer source disagrees with a secondary one, the more authoritative source is the displayed value; the secondary value is held internally only. The ProvenanceCard shows the source we displayed.

When two equally-authoritative sources disagree (e.g. an NPPES taxonomy code and a state-agency specialty designation), the field renders both with each one's source citation, not a synthesized merge. The reader gets to see the disagreement.

Edge case: a more recent state-level update disagrees with a slower federal snapshot. The federal snapshot continues to display, with its snapshot date attached, until the next reconciliation run resolves the disagreement.

Several source families produce fields we hold internally for matching, deduplication, or operator review but never expose on a public page — for example, NPPES sole-proprietor flags, internal entity-match keys, and dedupe fingerprints. The full list of write-locked-internal fields is documented in src/components/profile/source-family-copy.ts and surfaced in aggregate at /sources/[slug]'s display-policy block.

The display ceiling is a doctrine line, not a coverage gap. The §94 displayability rules are explicit: a field is rendered publicly only when (a) the source license permits it, (b) the field is doctrinally legible (no v-word adjectives, no fabricated quality stamps), and (c) the source's terms allow standalone display without ToS violation. Fields that fail any of those gates are held internally even when we have them.

Publisher cadence is a scheduling input, not proof that Fonteum has loaded the newest release. Re-ingestion can also be prompted when an authority changes methodology, a correction identifies stale data, an operator detects coverage drift, or a regulatory event makes a new observation a priority. Public pages report the loaded observation separately.

Retained row history is source-specific, not universal replay. Current retained-history rows exist for UK Companies House, federal procurement, selected provenance records, OFAC, and UN sanctions; other source families may expose only the loaded snapshot. The methodology changelog records material rule changes, not every historical row state.