Methodology

The assessment is built on DigComp 3.0, the European Digital Competence Framework for Citizens, published by the EU Joint Research Centre (Cosgrove and Cachia, JRC, November 2025; the fifth edition, after DigComp 2.2 by Vuorikari et al., 2022). DigComp also underlies the EU's digital skills indicator and the Europass digital skills self-assessment.

The assessment covers the five DigComp 3.0 competence areas (AI literacy and data privacy content is woven into each area rather than tested separately):

• Information and Data Literacy — browsing, searching, filtering, evaluating, and managing digital data and content
• Communication and Collaboration — interacting through digital technologies; sharing, engaging in citizenship, and collaborating
• Digital Content Creation — developing content, integrating and re-elaborating, copyright and licences, programming
• Safety and Wellbeing — protecting devices, privacy, health, and the environment; digital wellbeing
• Problem Solving — identifying needs, making decisions, resolving problems through digital tools; upskilling

Each area contains 12 questions, covering every competence in it. Every attempt draws a different set, preventing pattern memorisation. Three question types are used in each domain:

Scores within each domain are not revealed until the domain is complete — exam conditions apply throughout. The question bank is reviewed periodically against the current DigComp 3.0 specification.

There is no single pass mark. A level is awarded on two conditions at once: an overall score, and a minimum in every area sat — so strength in one area cannot pay for knowing nothing in another.Basic needs 50% overall and 40% per area, Intermediate 70% and 60%, and Advanced 85% and 75%. The names are DigComp 3.0’s own; Intermediate corresponds to the framework’s levels 3–4. Levels 7–8 (“Highly advanced”) are never awarded here, because they describe creating practice that a professional field adopts, which an unsupervised multiple-choice test cannot observe. These thresholds are provisional and set from published comparators; a defensible cut score requires a standard-setting panel and live statistics, which are still being collected.

A 24-hour cooldown between attempts discourages guessing and ensures that a passing result reflects genuine competence. Results are stored server-side; the certificate is issued only once, on the first passing attempt.

For psychometric detail — item statistics, reliability evidence, and standard-setting plans — see the Assessment Technical Manual.

The occupation you type is matched to an ESCO occupation and its ISCO-08 unit group (4 digits). If ESCO cannot be reached, the closest ISCO-08 group by title is used, and the page says so. For that group, two facts are looked up, never estimated:

• Generative AI exposure from the ILO’s 2025 index (Gmyrek et al., ILO Working Paper 140, Annex Table A1, CC BY 4.0). The ILO scored each occupation’s tasks and places the occupation in one of six categories from the average and spread of those scores: exposed at gradient 4, 3, 2 or 1, minimal exposure, or not exposed. The table covers 427 of the 436 ISCO-08 groups; for the other nine, no figure is shown.
• Green occupation: whether ESCO flags the occupation as green.

Exposure means the work may change, not that the job disappears. The ILO itself concludes that, because most occupations consist of tasks that need human input, transformation of jobs is the most likely impact.

The facts above, the ESCO description and ESCO skills are passed to a large language model (gpt-oss-120b, run by Groq). It writes the explanation of which tasks may change, a note on the green transition, five skills to build, a summary and advice. It is told not to change the facts, not to state any number or probability, not to forecast employment and not to predict job losses. The written parts may differ between runs.

Until October 2026 this page showed a 0–100 readiness score and automation-risk and AI-exposure percentages. Those came from a table TalentPath had written by occupation group, not from a published dataset, and were removed. There is no published method that turns these facts into one number, so none is shown.

Results are cached for four hours and describe the occupation, not the person.

Vacancy counts are retrieved in real time from national and international job board APIs. TalentPath does not estimate or model vacancy counts — every number shown is a live API response. Data sources by country priority:

For countries where a direct national API is available (DE, FR, SE), that source takes priority over aggregators. For all other member states, Adzuna is the primary source, with Jooble and EURES as fallbacks. Countries where no vacancy data is retrievable are shown as “no data source available” rather than zero.

The occupation title entered by the user is sent as a keyword query to each API. No mapping to ESCO codes is applied at the query stage — keyword search is used to maximise recall across national systems that do not all use ESCO occupation codes. Counts reflect the number of posted vacancies matching that keyword at the time of the request.

Results are cached for four hours server-side. Vacancy counts fluctuate daily — a result from earlier in the day may not reflect the current posting volume.

Skills entered by the user are mapped to ESCO v1.2(European Skills, Competences, Qualifications and Occupations taxonomy, European Commission 2022). ESCO contains over 13,000 skill concepts organised into a hierarchical taxonomy. Mapping uses semantic similarity search via sentence embeddings (HuggingFace models) against a locally indexed ESCO skills database, with fallback to the ESCO REST API (esco.ec.europa.eu).

Once skills are mapped to ESCO concepts, a large language model analyses the skill set against the target job title to produce:

• Demand signals — which skills are in growing, stable, or declining demand in EU labour markets, based on ESCO taxonomy positioning and sector trends
• Skill gaps — skills that are typically expected for the target role but not present in the user’s input
• Action plan — prioritised development recommendations

Important: Demand signals are LLM-assisted assessments grounded in the ESCO taxonomy, not live vacancy or salary data. They reflect the model’s analysis of structural skill demand patterns, not real-time labour market statistics. For real-time vacancy data, use the Skill Market tool.

• Assessment: weight 1
• Credential: weight 0.7
• Completed course: weight 0.3
• Self-report: weight 0. It never counts as evidence; it is kept only to compare how people rate themselves with what the evidence shows.

Manager ratings and anything inferred from work systems (e-mail, code, tickets) are not accepted. Evidence loses half its weight every 24 months by default (a skill can set its own half-life). Evidence older than two half-lives is “aged out” and asks for a fresh measurement. A person needs a combined weight of at least 0.6 before any reading is made for them.

Required levels are chosen from the four DigComp 3.0 proficiency steps, stored on the engine’s 0–1 scale:

• Basic (0.25): Carries out simple tasks, with guidance when needed.
• Intermediate (0.5): Carries out routine tasks and solves straightforward problems on their own.
• Advanced (0.75): Handles varied tasks and problems, chooses the right approach, and guides others.
• Highly advanced (1): Solves complex problems with few known solutions and improves how the field works.

Evidence from elsewhere is translated by fixed tables before it is stored. A TalentPath DigComp result reads as basic 0.25, intermediate 0.5 or advanced 0.75. An external credential counts according to how it was checked: accredited or signed as a credential, identified as a completed course (weight 0.3), declared as a self-report (no weight). A completed course evidences at most 0.5 (intermediate): it shows attendance, not mastery. The tables apply to evidence added from now on; rows already stored are read as they are.

Each person is estimated on their own evidence first. The group reading is the median person, with the middle half of the group as its range: a median, because the 0–1 scale orders people but its distances mean little. Nothing is shown for fewer than 5 people with evidence, so no reading ever describes one person. Confidence falls when the evidence covers less than 50% of the people concerned, and the page always says how many it covers.

The gap is a range: the required level minus the top and the bottom of the group’s range. A range that crosses zero reads as “can’t tell yet”, not as a gap. Readiness for an obligation is the share of each skill’s requirement that is met, weighted by how critical the skill is, and then capped at the weakest critical skill: two strong skills cannot hide a weak one. If a critical skill has no reading, readiness is not given at all.

• Measure first: there is not enough evidence, or it has aged out.
• Recognise: there is no real gap, or people rate themselves below what the evidence shows.
• Hire: a real gap, and the market can supply the skill (your fillability setting 0.6 or above).
• Build: a real gap the market cannot supply (0.3 or below).
• Cover in the interim: as for build, but urgent.
• Build and hire: a real gap in a tight market, in between.

Skills are ranked, never people: the size of the gap × how critical the skill is × how many people it reaches, raised when the market cannot fill it and divided by the effort to close it. Each recommendation lists its reasons on the page.

The user’s occupation and skills are mapped to the ESCO v1.2 occupation taxonomy using the same semantic search approach as Skill Intelligence. ESCO occupations are linked to ISCO-08 codes, enabling cross-border comparability since ISCO-08 is the international classification used by EU employment services, Europass, and EURES.

For regulated professions, portability analysis references Directive 2005/36/EC on the recognition of professional qualifications (as amended by Directive 2013/55/EU), which governs recognition of professional qualifications between EU member states. Three recognition types apply:

• Automatic recognition — for professions with harmonised minimum training requirements (doctors, nurses, pharmacists, architects, etc.)
• General system recognition — competent authority comparison, with possible compensation measures
• Partial access — available under specific conditions where the professional activity can be separated from the full profession

Regulatory data is updated periodically from the European Commission’s regulated professions database. Portability guidance on this platform is indicative only. Users must verify current requirements with the relevant competent authority in the target country. Requirements change and this tool does not constitute legal advice.

Where EQF level information is available via ESCO, it is used to contextualise cross-border qualification recognition. EQF levels 1–8 provide a reference for comparing qualifications across different national systems, supporting the transparency objective of the Skills Portability Initiative (European Commission, 2023).

Certificates issued by TalentPath are W3C Verifiable Credentials (W3C Recommendation, March 2022, updated 2024) structured as JSON-LD documents. The credential payload uses the European Learning Model v3 (ELM v3)JSON-LD context, the same schema used by Europass for digital credentials.

Each of the five competence-area badges is a micro-credential — a small-volume, verifiable record of learning outcomes that stacks with the others into the full Digital Skills Credential. This design is aligned with the EU's micro-credential approach (Council Recommendation of 16 June 2022 on a European approach to micro-credentials for lifelong learning and employability); TalentPath badges are aligned with that approach, not officially recognised qualifications.

Each credential is cryptographically signed using Ed25519Signature2020, an elliptic curve digital signature algorithm providing 128-bit security. TalentPath operates as the credential issuer under its own DID (Decentralised Identifier), published at /.well-known/did.json. The signing key pair is generated and stored securely; the private key never leaves the server environment.

Once issued, a credential’s content cannot be altered retroactively — any change breaks its signature. If a credential must be revoked, TalentPath marks it revoked at the issuer; verification at talentpath.eu/verify then reports it as no longer valid, while the credential document itself remains on the user’s device with its original signature intact. Each credential also carries a W3C Bitstring Status List entry, so any third party can check revocation offline by fetching the published status list — no need to contact TalentPath.

Any certificate can be verified at talentpath.eu/verify by entering the certificate code. Verification confirms: the credential was issued by TalentPath, the content has not been tampered with, and the credential has not been revoked. Verification exposes the holder’s name, overall score, domain breakdown, and issue date — the holder consents to this by completing the assessment.

TalentPath implements the OpenID for Verifiable Credentials (OID4VC) issuance protocol, so a credential can be offered to a wallet app that supports it. The endpoint is live at/api/oid4vc/credential. This is not EUDI Wallet compatibility, and TalentPath does not claim it: that also requires an SD-JWT VC or ISO mdoc encoding, the wallet’s interoperability profile (HAIP), and registration as a trusted issuer. An SD-JWT VC export is the first step on the roadmap.