Methodology

The Robotics Census publishes a company only when recorded evidence proves what it does. This page states exactly how that works, what the census holds, and why it publishes so much less than it tracks. The honesty is the point: an unevidenced claim on a public page is the one thing this census exists to avoid.

Where the data comes from

The census is built from open, citable sources, not from scraped directories or paid data brokers:

• EU CORDIS open data. The European Commission's open dataset of funded research participants: legal name, registered identifier, address, coordinates, project participation and EU grant amounts. Every fact traces to a public record.
• A3 member roster. The public membership listing of the Association for Advancing Automation, including the technology tags a member declares for itself.

Where a company has a working website, the census also reads its homepage directly and records the exact text it relied on. That text, verbatim, appears on the company's profile.

The pipeline, end to end

Five steps take a raw dataset to a published page, and every step leaves an audit trail:

• Acquire. Raw payloads are downloaded and recorded in a manifest with URL, checksum, size and fetch time, so any run can be verified byte for byte.
• Normalize. Entities are extracted with a stable identity rule. A company's registered identifier anchors identity where one exists; otherwise a normalized name plus country does. A web domain is an attribute of a company, never a basis for merging records, and name similarity alone never merges anything.
• Track. Every entity is loaded with provenance rows naming the dataset, source URL and fetch date it came from.
• Assess. Companies are classified core, adjacent, participant-only or unrelated by a deterministic procedure over recorded evidence. Decisive signals are read only from a page's declared identity zones, its title, description and headings, never from marketing copy in the body. A language model's opinion is never treated as evidence. Assessments below high confidence go to editorial review instead of being recorded as fact.
• Publish. The site is regenerated from the database at build time. Only core and adjacent companies are published, and a page renders no fact without a source URL and an as-of date behind it.

What the census holds, and what it publishes

The census tracks 5,937 companies and 2,501 institutions in total. It publishes 509 companies. The gap is deliberate.

A company is published only when its own website or an association technology tag evidences what it does. The rest are tracked but not published, because the census has nothing it can prove about them yet. Current per-country and per-category tables are on the statistics page; underlying data as of 2026-07-15.

The four classes

Every company the census can assess falls into one of four classes. Only the first two are published.

Core

Designs, builds or sells robots or robotic systems. Published.

Adjacent

Supplies components, sensors, actuators, drives, software, simulation, integration or services specifically for robotics or industrial automation. Published.

Participant only

The actual business is something else, and robotics was only the context: a pilot site, an end user, a data provider, a consortium partner. Not published. A hospital that hosted a robotics trial is not a robotics company.

Unassessed

The census has no usable evidence either way, so it makes no claim and creates no record. Not published.

Core versus adjacent tests what a company sells. Adjacent versus participant-only tests whether robotics is the business or merely the context. The terms themselves are defined in the glossary.

Project participation is never enough

Appearing in an EU-funded robotics project does not make a company a robotics company. Banks, packaging firms, hospitals and consultancies all join robotics consortia. The census treats project participation as context, never as proof. A company is published only when its own declared identity, on its website or in an association tag it chose, shows what it does.

Absence is not a judgment

If a company is not in the census, that is not a verdict about the company. It means the census does not yet hold evidence it can cite. The census would rather omit a real robotics company than publish an unproven claim about one.

Getting listed or corrected

To be listed, or to correct something on an existing page, email francesco@bianchiweb.com with a link to the evidence: your website, or a public record showing what the correction should be. Corrections are applied as an editorial layer with the evidence recorded alongside them.

Frequently asked questions

How does a company get into the Robotics Census?

A company is published when recorded evidence shows what it does: its own website, read directly with the exact text recorded, or a technology tag it declared for itself on the A3 member roster. Companies enter tracking through the EU CORDIS open dataset and the A3 roster, but tracking alone never publishes a company.

What does core mean in the census?

A core company designs, builds or sells robots or robotic systems. The evidence for that assessment is quoted verbatim on the company profile with its source URL and date.

What does robotics-adjacent mean in the census?

An adjacent company supplies components, sensors, actuators, drives, software, simulation, integration or services specifically for robotics or industrial automation. It serves the robotics industry without necessarily building robots.

Why is a robotics company I know missing from the census?

Absence is not a judgment. It means the census does not yet hold evidence it can cite: the company may not appear in the source datasets, or its evidence has not been assessed yet. The census would rather omit a real robotics company than publish an unproven claim about one.

What data sources does the Robotics Census use?

Three: EU CORDIS open data on funded-research participants, the public member roster of A3 (the Association for Advancing Automation) with its declared technology tags, and company websites read directly. Every profile lists the datasets it came from with fetch dates.

Is participating in an EU robotics project enough to be listed?

No. Project participation is treated as context, never as proof. Banks, hospitals and packaging firms all join robotics consortia; participation alone leaves a company unpublished or assessed participant-only.

How often is the census updated?

Every page is regenerated from the census database whenever the site is built, and each rendered fact carries the fetch date of its source. Source datasets are refreshed in recorded, dated sweeps rather than on a fixed calendar; the manifest of raw payloads records the URL, checksum and fetch time of every ingested dataset version.

How do I correct an error on a census page?

Email francesco@bianchiweb.com with a link to the evidence: your website, or a public record showing what the correction should be. Corrections are applied as an editorial layer with the evidence recorded alongside them.