Same label, different meaning
An identical label can describe different tasks, performance levels and evidence.
Published working draft · Invitation to review
The verified vector graph for behaviour anchors and competencies.
A skill label alone is not evidence of competence. The nonprofit OSC describes competencies through behaviour anchors – observable action, context, level and admissible evidence – finds equivalents across languages with vectors and approves every match only after expert review.
- Nonprofit and open
- Human-in-the-loop approval
- Built on peer-reviewed research
- BasicIdentifies documented error codes and performs a prescribed test step correctly.
- AdvancedPrioritises several fault hypotheses from measurement data, tests them systematically and documents the decision path.
- ExpertDevelops a cross-system test plan, rules out alternative causes and turns findings into improved diagnostic procedures.
Illustrative example from the published OSC working draft
The problem
A skill label alone is not evidence of competence.
“Problem solving”, “data literacy” or “teamwork” appear in job profiles, curricula and competency models worldwide – yet nobody can be sure that different organisations, sectors or countries mean the same thing.
Different words, same behaviour
Differently worded skills can describe almost the same behaviour in practice – and keyword matching misses them.
What sets the OSC graph apart
Behaviour instead of buzzwords. Vectors propose, experts decide.
Behaviour anchors
Every skill is tied to observable action, application context, performance level and admissible evidence – not just a name.
Multilingual vector layer
Embeddings find semantically similar descriptions, translations and duplicates across systems and languages – as candidates for a match.
Expert approval with provenance
Domain owners review proposals against defined criteria. Approvals, rejections, dissent and evidence are stored as auditable provenance.
Answers with sources
A GraphRAG-ready access layer answers from relevant subgraphs and points back to the nodes, edges and evidence it used.
Global means verifiable
A shared open core with controlled local variants: documented equivalences and differences, separate approvals per domain, language and context.
“Vector similarity produces a proposed match – not semantic truth.”
Architecture
Five layers from skill hypothesis to approved knowledge.
The binding meaning stays in the versioned knowledge graph – with explicit relations, sources, scopes and approval status. Machines scale search, clustering and proposals; qualified experts decide on meaning, level, context and approval.
Explore the graph model- 01Ontology & knowledge graph
Skills, tasks, roles, qualifications, learning offers, evidence, anchors and sources with stable identities and typed, versioned relations.
- 02Multilingual vector & alignment layer
Similar descriptions, translation variants and duplicates as match candidates – for crosswalks between existing taxonomies.
- 03Behaviour anchor & evidence layer
Observable actions, contexts, performance levels and admissible types of evidence.
- 04Expert validation & governance
Defined review criteria; high-impact decisions never rest on an unchecked LLM rating alone.
- 05GraphRAG access & reasoning
Questions answered from subgraphs, relations and sources – with references back to them.
- Workflow event
- Skill hypothesis
- Human review
- Learning intervention
- Outcome
Who it is for
One verifiable basis for the whole skills ecosystem.
Employers
Link role profiles to concrete tasks, evidence and learning paths.
Education providers
Relate learning outcomes precisely to observable professional action.
Chambers & recognition bodies
Show similarities and differences between qualifications – without automatically claiming equivalence.
Labour-market institutions
Spot new skill patterns earlier and review their classification transparently.
Employees & learners
Portable evidence that carries context and proof, not just a label.
HR tech, LMS & talent marketplaces
Versioned, machine-readable and source-linked skill relations.
Research foundation
Built on peer-reviewed research – and open to review itself.
Behaviour-based competency description, expert validation, multilingual embeddings, knowledge-graph alignment and graph-based retrieval are well supported in the literature. The concrete OSC architecture is a published working draft and has not yet been peer-reviewed as a whole system – that is exactly why we are inviting experts now.
Read the full working draft (German)- Behaviour AnchorsSmith & Kendall (1963), Journal of Applied Psychology
- Competency modellingCampion et al. (2011), Personnel Psychology
- Expert validationLievens & Sanchez (2007), Journal of Applied Psychology
- Graph embeddingsBordes et al. (2013), NeurIPS
- Multilingual embeddingsReimers & Gurevych (2020), EMNLP
- Skill extractionZhang et al. (2022), NAACL
- GraphRAGHe et al. (2024), NeurIPS
- Measurement invarianceVandenberg & Lance (2000), Organizational Research Methods
Get involved
Review anchors, test the graph, shape the standard.
We are looking for experts in vocational education, competency assessment, work psychology, recognition, ontology engineering, knowledge graphs, multilingual AI and labour-market research – and organisations with real skills data for a pilot.
Frequently asked
Open Skills Consortium in brief.
What is the Open Skills Consortium (OSC)?
The Open Skills Consortium is a nonprofit organization (AISBL in formation). It develops an open vector graph that describes competencies through behaviour anchors and approves every match only after expert review – as a global digital public good.
What is a behaviour anchor?
A behaviour anchor describes the observable action, application context, performance level and admissible evidence of a skill. The idea goes back to behaviourally anchored rating scales (Smith & Kendall, 1963); OSC uses it as a semantic layer of its skills ontology.
What does “verified” mean in the OSC graph?
Vector similarity only produces proposed matches, not semantic truth. Domain owners review every proposal against defined criteria; approvals, rejections, dissent, evidence, language version and scope are stored as auditable provenance.
Why a vector graph instead of a competency list?
The knowledge graph holds the binding meaning with explicit relations, sources and approval status. The multilingual vector layer finds similar descriptions, translations and duplicates across systems and languages. Together, skills become comparable without confusing similarity with equivalence.
Has the approach already been validated?
Its building blocks are well supported by peer-reviewed research. The concrete OSC architecture is a published working draft and has not yet been peer-reviewed as a whole or validated worldwide. A pilot therefore tests five paths: semantic alignment, anchor quality, language and cultural validation, graph and provenance quality, and GraphRAG evaluation.
How can I get involved?
Experts can review behaviour anchors, organisations can test the graph in a pilot with their own data or become members. The form on this page is enough as a first step.
Not the next closed skills list – an open, verifiable graph.
Help decide which behaviour anchors can be referenced globally, where local variants are needed and which evidence is enough for approval. Nonprofit, AISBL in formation.