Where do the evidence values come from?

From FDA's own published decision and 510(k) summaries. Each extracted value is shown with its document, page number and excerpt, and is marked review recommended rather than presented as verified fact.

What happens when a metric is not in the document?

The field reads Not found in public documentation. It is never left blank, and it is never filled with an inferred value.

AI / SaMD evidence benchmark

AI and SaMD evidence benchmark, extracted from FDA decision summaries

FDA's AI-enabled device list tells you which devices were authorized. It does not tell you what evidence supported them. That sits inside decision summaries, as PDFs, one device at a time.

Who it is for: AI/ML and digital health teams building an evidence strategy, and reviewers benchmarking a planned study against what has already been authorized.

Start a 14-day free trial — no card needed See plans

What is in the workspace today

AI-enabled devices

Count unavailable

Reading the latest refresh…

Side-by-side comparison, with provenance

Compare two to ten devices on study design, cohort size, reader design, external validation, ground truth and reported performance metrics. Every value carries the document it came from, the page number and the excerpt, so a reviewer can check the source in one click.

Documents keep their FDA file link, FDA page link, per-page numbering and a SHA-256 checksum
Revision history is retained, so a document FDA republishes does not overwrite what you reviewed
Similar-device discovery surfaces comparable authorizations by product code and evidence shape

Absence is a stated finding

A field the documents do not contain reads Not found in public documentation. A document FDA's site would not serve is retried on a widening schedule, and a read failure is never recorded as no document published — that stays a separate, dated finding, re-confirmed after 30 days.

Where the data comes from

Only published regulator records, retrieved from official sources. Every screen links back to the record it came from.

FDA AI-enabled medical device list
Authorized AI-enabled devices with their submission numbers.
Open the regulator source
FDA decision summaries and 510(k) summaries
The published documents the evidence fields are read from.
Open the regulator source

What this module does not do

Values are machine-extracted from FDA documents and marked review recommended. Confirm anything you rely on against the linked document and page.
Extraction covers a growing subset of the FDA AI-enabled device list; the page states how many devices have documents read and how many are queued.

Questions

Where do the evidence values come from?: From FDA's own published decision and 510(k) summaries. Each extracted value is shown with its document, page number and excerpt, and is marked review recommended rather than presented as verified fact.
What happens when a metric is not in the document?: The field reads Not found in public documentation. It is never left blank, and it is never filled with an inferred value.

Related modules

Market intelligence: FDA ↔ Canada

Compare FDA 510(k), De Novo and PMA authorizations against Health Canada MDALL licences with scored company and device matching, and find the US devices with no apparent Canadian counterpart.

QMSR enforcement

FDA inspection citations, published observations and medical device warning letters, classified Pre-QMSR or QMSR-era, mapped to a quality-system taxonomy and linked to the companies named.

All six modules are in the platform overview, and every plan includes the full source history — see pricing.