Germany’s AI innovation path

What public evidence proves—and where the link ends

The evidence path connects German EU funding participation to reported research outputs through exact CORDIS identifiers. Patents and procurement remain visibly separate until a shared official identifier proves the link.

  • Exact project and PIC identifiers
  • Uncertain links excluded
  • Deterministic · no generator
Bounded, exactly linked starting cohortProjection dated 14 Oct 2026, 05:17Observed since 13 Jul 2026CORDIS funding data not refreshed since 29 Sept 2026
German net EU allocation€1.23bn

German participation only within the exactly linked CORDIS projects

Proven project paths1,156

CORDIS project code and participant PIC agree in both sources

Project-output relations57,360

OpenAIRE relations, not claims of authorship or impact

Citations across project relations799,557

Sum of OpenAIRE citation counts per project: an output of several projects counts once for each. Not a quality or impact score.

Evidence-path explorer

Follow every proven step

Choose a project and inspect its identifiers, funding amount, research outputs and the deliberately open transitions to patents and public procurement.

1,156 paths
H2020 · 101000165

leArning and robuSt deciSIon SupporT systems for agile mANufacTuring environments

01 Nov 2020 to 31 Oct 2023

CORDIS project ↗
  1. 01proven

    EU allocation

    German project participation

    €1.45m

    Reported net EU contribution to German participants in this CORDIS project.

  2. 02proven

    Participating organisation

    4 German participants

    PIC 897164835

    Role reported by CORDIS: participant.

  3. 03proven

    Reported research

    ASSISTANT

    82 outputs

    73 with open access; 513 reported citations.

  4. 04not linked

    Patent families

    Separate AI patent monitor with observed German signals

    20,344 families · not linked

    The EPO monitor is collected independently. No family is automatically assigned to this CORDIS project or its participants.

  5. 05not linked

    Public procurement

    Tender to contract

    No exact bridge

    OCDS and TED party identifiers cannot currently be equated safely with CORDIS PICs.

EPO OPS · automatically observed

AI patent monitor

The monitor counts published records that match the fixed, deliberately narrow CPC rule for core AI in EPO OPS. It shows observed publication activity—not current inventions, patent quality or a connection to the selected funding award.

Covered search window completePublication dates from 29 May 2026 to 30 Sept 2026 have been processed completely for this bounded search window. This confirms coverage of that window, not historical completeness of the monitor.Data dated 14 Oct 2026, 04:23
Observed DOCDB family IDs20,344

The counting unit is a deduplicated DOCDB simple-family ID, not each individual publication.

Observed search publications23,627

Only publications returned by the fixed OPS search.

Automatically observed since16 Jul 2026

The start of data collection—not the start of patent activity.

Observed family IDs by publication month

Only fully covered calendar months are compared as a trend. Partial months remain visible as non-comparable raw values; assignment follows the earliest publication date observed for a family ID by the search run.

Observed monthly values with explicit coverage status
MonthFamily IDsCoverage
May 20263,417Partial month · not comparable
Jun 20264,302Full month · comparable
Jul 20263,569Full month · comparable
Aug 20266,044Full month · comparable
Sept 20263,012Full month · comparable

Partial months are excluded from the trend bars and must not be compared with one another or with full months.

Observed family IDs by fixed CPC rule

The deliberately narrow, versioned rule classifies the family IDs observed in the search window in a traceable way.

  • Neural networks15,421
  • Machine learning7,533
  • Knowledge-based models5,523
  • Probabilistic graphical models1,379
  • Legacy AI classification60
  • Fuzzy logic53

Categories overlap: one patent family may be counted in more than one bar.

Observed German signals—not an origin claim

Each dimension answers a different question. The values must not be added or read as an organisation link. Zero only means that no such evidence was observed in the returned OPS records—not that the association is ruled out. Residence and address fields may be incomplete.

Applicant residence or seat country: DE
not captured
Inventor residence or address country: DE
not captured
German priority filing
240 families
German publication
190 families

This run has no usable residence data for applicants or inventors. These two values are therefore unknown, not zero.

Boolean union: at least one German association334

Each family is counted at most once here. The union is not a fifth evidence dimension.

Observed publication countries and authorities

Breakdown of publications observed in the search result. EP and WO are publication authorities, not countries of residence or incorporation.

  • China · CN8,847 families · 8,871 publications
  • United States · US6,640 families · 6,766 publications
  • WIPO · WO2,738 families · 2,738 publications
  • European Patent Office · EP2,146 families · 2,158 publications
  • Japan · JP995 families · 1,005 publications
  • South Korea · KR972 families · 982 publications
  • Australia · AU220 families · 222 publications
  • Germany · DE190 families · 191 publications
  • Spain · ES141 families · 145 publications
  • United Kingdom · GB110 families · 110 publications
  • Canada · CA75 families · 75 publications
  • France · FR59 families · 59 publications

The twelve countries or authorities with the most observed search publications are shown.

Newest observed family IDs

A bounded, deterministic sample without titles, abstracts, person names or organisation names.

Showing 12 of 20,344 families eligible for the public sample.

  1. DOCDB 101490442

    1 observed publication in one country or publication authority

    Categories
    Neural networks
    CPC symbols
    G06N3/045 · G06N3/088
    Publication countries and authorities
    GB
    German associations
    no observed German association
  2. DOCDB 64051830

    1 observed publication in one country or publication authority

    Categories
    Machine learning · Neural networks · Probabilistic graphical models
    CPC symbols
    G06N20/00 · G06N20/20 · G06N3/002 · G06N3/02 · G06N3/04 · G06N3/044 · G06N3/045 · G06N3/0464 · G06N3/048 · G06N3/061 · G06N3/08 · G06N3/082 · G06N3/084 · G06N3/086 · G06N3/0895 · G06N3/09 · G06N3/123 · G06N7/01
    Publication countries and authorities
    ZA
    German associations
    no observed German association
  3. DOCDB 69529024

    1 observed publication in one country or publication authority

    Categories
    Neural networks
    CPC symbols
    G06N3/04 · G06N3/0464 · G06N3/08 · G06N3/09
    Publication countries and authorities
    EP
    German associations
    no observed German association
  4. DOCDB 69811938

    1 observed publication in one country or publication authority

    Categories
    Knowledge-based models
    CPC symbols
    G06N5/01
    Publication countries and authorities
    EP
    German associations
    no observed German association
  5. DOCDB 74041737

    1 observed publication in one country or publication authority

    Categories
    Knowledge-based models · Neural networks · Probabilistic graphical models
    CPC symbols
    G06N3/04 · G06N3/044 · G06N3/045 · G06N3/0455 · G06N3/0464 · G06N3/048 · G06N3/08 · G06N3/084 · G06N3/09 · G06N5/046 · G06N7/01
    Publication countries and authorities
    ZA
    German associations
    no observed German association
  6. DOCDB 75919334

    1 observed publication in one country or publication authority

    Categories
    Machine learning · Neural networks · Probabilistic graphical models
    CPC symbols
    G06N20/00 · G06N20/20 · G06N3/08 · G06N3/0895 · G06N7/01
    Publication countries and authorities
    ZA
    German associations
    no observed German association
  7. DOCDB 84569812

    1 observed publication in one country or publication authority

    Categories
    Machine learning · Neural networks
    CPC symbols
    G06N20/00 · G06N3/045 · G06N3/0464 · G06N3/088 · G06N3/0895 · G06N3/09 · G06N3/096
    Publication countries and authorities
    ZA
    German associations
    no observed German association
  8. DOCDB 84818659

    1 observed publication in one country or publication authority

    Categories
    Machine learning · Neural networks
    CPC symbols
    G06N20/00 · G06N3/0455 · G06N3/0464 · G06N3/088
    Publication countries and authorities
    ZA
    German associations
    no observed German association
  9. DOCDB 86903069

    1 observed publication in one country or publication authority

    Categories
    Neural networks
    CPC symbols
    G06N3/065 · G06N3/08 · G06N3/09
    Publication countries and authorities
    ES
    German associations
    no observed German association
  10. DOCDB 88874742

    1 observed publication in one country or publication authority

    Categories
    Neural networks
    CPC symbols
    G06N3/02 · G06N3/045
    Publication countries and authorities
    EP
    German associations
    no observed German association
  11. DOCDB 88923786

    1 observed publication in one country or publication authority

    Categories
    Neural networks
    CPC symbols
    G06N3/08
    Publication countries and authorities
    EP
    German associations
    no observed German association
  12. DOCDB 88969458

    1 observed publication in one country or publication authority

    Categories
    Machine learning
    CPC symbols
    G06N20/00
    Publication countries and authorities
    EP
    German associations
    no observed German association
How to read the AI patent monitor
  • The counting unit is the DOCDB simple-family ID; the same ID is not counted anew for every observed publication.
  • Selection follows the fixed CPC rule version shown below. Historical restatement with later rule versions is not currently available.
  • Observed search publications are not a complete list of all members of a patent family.
  • The monitor measures observed publication activity; it does not assess invention timing, patent quality, legal status or funding impact.
  • Collection, classification and projection do not use a generative AI model.

Taxonomy epo-cpc-ai-high-precision · version 2026.01

Evidence register

Every edge carries its proof

Chronology is not causality. The register therefore shows which relation is exactly proven and which remains open because no shared identifier exists.

StatusRelationLink basisSource
provenFunding record belongs to the selected projectCORDIS project code 101000165CORDIS ↗
provenGerman participant belongs to the allocation rowCORDIS PIC 897164835CORDIS ↗
provenOpenAIRE products are reported against the projectOpenAIRE project ID corda__h2020::cb79172f2c32c987b4e5452b1d75d118 + CORDIS 101000165OpenAIRE ↗
openPatent family belongs to project or organisationEPO supplies no stable applicant identifier for a cross-source joinEPO OPS ↗
openProcurement belongs to project or organisationNo shared official identifier between PIC and procurement partyOCDS / TED ↗
Gaps are data

What we explicitly refuse to guess

The path becomes more valuable because missing bridges stay visible. A name, similar address or shared region is not enough for an organisation or impact claim.

01 · Patents

Applicant names are not global IDs

DOCDB patent families are collected as a separate index. We only publish a link to CORDIS, ROR or LEI when a source-backed shared identifier exists.

02 · Procurement

Source-local parties remain separate

TED and OCDS identifiers connect lifecycles within their source. They are not merged with funded organisations through name similarity.

03 · Impact

Chronology does not prove cause

Funding before an output, patent or contract is a time pattern. The index does not label it commercialisation or funding impact.

Method & identity rules

A conservative data spine

The public path is built only from relations that both participating sources support with compatible identifiers.

  1. 1
    Intersect exact project codes

    An allocation row enters the path only when its CORDIS project code exactly matches the transfer project.

  2. 2
    Confirm participants through PIC

    The German organisation must carry the same CORDIS PIC in funding and transfer data. Names are display values, never join keys.

  3. 3
    Check output relations twice

    The OpenAIRE project ID and CORDIS grant identifier must both occur in the product relation.

  4. 4
    Preserve disconnected lanes

    Patent and procurement data keep their own source identities. Candidates, fuzzy matches and private review data never reach the website.

Collection, classification and linking do not invoke a generative AI model.

Sources & coverage

The first public revision is deliberately bounded. Every source entry discloses its own scope; missing data is never reinterpreted as zero.

European Commission CORDISfully checked
1,156 records

Bounded transfer probe with exact grant identifiers and German participant PICs.

checked 01 Oct 2026, 07:17Open official source ↗
OpenAIRE Graph funded-products datasetfully checked
55,858 records

Reported research products and project relations; the public view is bounded while project aggregates come from the fully checked run.

checked 01 Oct 2026, 07:17Open official source ↗
European Commission CORDISfully checked
2,551 records

Official Horizon Europe bulk data underlying the funding and participant records.

checked 29 Sept 2026, 05:17Open official source ↗
European Commission CORDISfully checked
2,364 records

Official Horizon 2020 bulk data underlying the funding and participant records.

checked 29 Sept 2026, 05:17Open official source ↗
European Patent Office · OPSavailable · not linked
20,344 records

Automatically collected DOCDB patent families from EPO OPS. The lane is available but remains separate from CORDIS projects and organisations without a shared official identifier.

checked 14 Oct 2026, 04:23Open official source ↗
Datenservice Öffentlicher Einkauf · TEDnot linked

Procurement lifecycles are already collected daily. A source-backed shared identifier with CORDIS participants is not currently available.

Open official source ↗

Frequently asked questions

Does the path show what funding caused?
No. It shows reported allocations, project participation and research-product relations in chronological order. This proves neither causality nor economic impact.
Why not connect organisations by name?
Legal forms, spelling, corporate groups and subsidiaries make names ambiguous. A false merge could assign money, patents or contracts to the wrong organisation. We therefore require compatible source-backed identifiers.
Why are patents and procurement not shown as zero?
Not linked does not mean none exist. Without a safe identity bridge, the project-level count is unknown, so we deliberately avoid showing zero.
What does the separate AI patent monitor capture?
It counts published EPO OPS records that match the fixed, deliberately narrow CPC rule for core AI, deduplicates them by DOCDB family ID and reports observed German signals separately. It measures publication activity and does not automatically connect any family to a CORDIS project.
Does an OpenAIRE relation prove German authorship?
No. CORDIS proves German project participation; OpenAIRE proves a product relation to the project. Without a separate exact organisation identifier, we make no authorship claim.
Does the collector use generative AI?
No. The complete path uses versioned rules, exact identifiers, schema validation and traceable source evidence.
Traceable data products

Want to connect data sources without inventing evidence?

We build robust collectors, archives and analysis pipelines with clear identity rules and verifiable provenance.

Discuss a projectView all tools