AI grounding
A model can cite an answer only if the answer carries a source trail. Fonteum makes the trail travel with the fact: source name, source URL, capture date, limitation, hash, and signature.
Fonteum · Manifesto
Most data tools ask you to trust the answer. Fonteum hands you the government record: dated, sourced, signed, attested, chained, and re-checkable, including what it used to say.
Last updated 2026-06-24 · Aggregate public-records positioning · FCRA-safe by design
44
federal source families
Registry-derived, not hand-entered into this page.
9M+
providers in the broad NPPES identity spine
An aggregate public-records population, not a provider rating.
11.9M
historical NPPES and PECOS source rows
Used as data-volume context; sources are counted separately.
The thesis
The market has learned to love summaries. Dashboards summarize. Agents summarize. Search results summarize. The risk is that a summary can feel authoritative while separating the reader from the public record underneath it.
Fonteum is built for the opposite instinct. A number should carry where it came from. A field should carry when it was captured. A conclusion should carry its limitation. A record should be re-checkable by someone who did not build the system.
The answer is useful only when the source can follow it into the room.
That is not a UX preference. It is an operating position. If a fact is going to ground an AI agent, support a compliance review, or sit under a published finding, the fact needs more than a label. It needs a trail.
What travels with a fact
Fonteum treats provenance as payload, not as decoration. The record should say what agency published it, which dataset it came from, when Fonteum captured it, which methodology touched it, and what limitation belongs beside it.
These words describe the record trail. They are not provider endorsements.
Why now
AI systems need grounding that survives citation. Compliance teams need evidence that survives audit. Journalists and researchers need figures that readers can retrace. Three different audiences, one shared failure mode: an answer with no source trail becomes a dead end.
A model can cite an answer only if the answer carries a source trail. Fonteum makes the trail travel with the fact: source name, source URL, capture date, limitation, hash, and signature.
A compliance file needs more than a current answer. It needs the record as of the day the decision was made, plus enough context for an auditor to re-run the check.
A public-interest finding is stronger when readers can inspect the same government file. The work should be reproducible from the source, not dependent on a screenshot or a vendor memo.
The time-machine edge
Public datasets change. A provider updates an address. An enrollment appears or disappears. A sanctions file is refreshed. A facility measure is republished. If a system only stores the current value, it cannot answer the question that matters later: what did the source say when the decision was made?
Fonteum is not just a lookup table. It is a dated memory of public records.
Every snapshot creates an as-of layer. The older value is not erased from the narrative; it becomes part of the chain. That is the edge a fresh scrape cannot backfill after the fact.
The stance
Fonteum is deliberately narrow about what belongs in public copy and customer-facing data. If a source cannot be named, if a date cannot be shown, if a limitation would change how the reader should use the field, the system should say so.
The line
Fonteum exists for the moment after someone asks, "Where did that come from?" The answer should be immediate: from this government record, captured on this date, handled by this method, signed into this chain, and re-checkable from here.
Built on the authoritative federal record
The primary sources, named on every page.
These are the federal agencies whose public datasets Fonteum ingests and attributes — the issuing authorities, not customers or partners. Every figure on the site links back to one of them.
See the full source registry, with license and refresh cadence for each →
Reproducible by design
Every figure traces to its federal source.
14-tuple provenance
Every rendered fact ties to a source URL, dataset ID, snapshot date, row key, and SHA-256 — the full chain-of-custody record.
Reproducible SQL
Each study ships the exact query behind its figures, run against the cited federal snapshot. Re-run it yourself.
Daily count checks
Published counts are checked against the upstream federal datasets on a daily cadence, with drift logged.
Named medical review
Reviewed by Jennifer Montecillo, MD, medical reviewer. Non-practicing medical reviewer.
Two doors
Use the free API and open data
Query providers, facilities, sanctions, and quality scores — each field carrying its federal source. Self-serve, no call to start.
Talk to us
Managed pilots, enterprise terms, and audit-ready, signed attestation packages for compliance, risk, and research teams.
The substrate, by the numbers