Procedure manual and professional information

Purpose and evidence status

For people exploring study options. Results are starting points for discussion, not an aptitude diagnosis, an admissions assessment or a prediction of academic success. There are no empirically established age norms, diagnostic cut-offs or demonstrated equivalence to established counselling instruments.

• Theory: RIASEC interests and Mini-IPIP personality items. Research on these models does not automatically validate our items, translations, field profiles or rankings.
• Evidence for the complete procedure: reliability, test–retest stability, construct and criterion validity have not been demonstrated in our own published studies. Software tests check behaviour, not psychometric quality.
• Norms: there is no documented representative German-speaking norm sample. We therefore report no age comparisons or personality percentiles.
• Practical approach: explore programmes, school subjects, interests and optional everyday topics together. Added counselling value over established instruments remains a hypothesis to test.

Administration

1. Explain the purpose, lack of validation and voluntary data sources before starting. Do not require personal exports on an adviser’s or shared device. Use fictitious responses for demonstrations.
2. Respond independently without steering towards a desired result. Interests: 1 = dislike to 5 = love. Personality: 1 = very inaccurate to 5 = very accurate, considering typical behaviour. There is no time limit; standardised completion times have not been measured.
3. 24 interest items and one attention item; 20 personality items; school-subject and work-value self-reports. A result requires at least two interest answers per type or one imported source. Personality is scored only after 20 valid responses.
4. Treat questionnaire, source analysis and AI comparison as different conditions. For a repeat, keep language, version and data basis consistent; document changes and interruptions. Review results together; the person makes their own decision.

Scoring and interpretation

Interests: each RIASEC response mean is mapped to 0–1 using (mean − 1) / 4. At least two of four items per type are required; missing types are not treated as neutral. The six scores and three highest types describe the interest profile, not a diagnostic classification.

Personality: only integer responses from 1 to 5 count. Negative items are reversed with 6 − response; the mean of all four items is displayed for each trait. Missing responses are not imputed. Example extraversion: b1=5, b6=1, b11=4, b16=2 gives (5+5+4+4)/4 = 4.50 out of 5. This is not a percentile, age comparison or clinical finding. Internally the mean is centred on the response midpoint of 3.

Field ranking with an interest questionnaire: available components start at 50% interests, 36% school subjects, 14% values and 0% personality. Unavailable components are omitted and the rest reweighted. Interest fit combines profile shape (70%) and distance (30%). School subjects combine enjoyment (60%) and self-rated performance (40%). Optional sources contribute at most 25% of overall fit. These weights are product decisions, not empirical calibrations.

Without a scorable interest questionnaire, the ranking is exploratory and uses available signals. Scores are relative to fields in that result and are not directly comparable with questionnaire-mode scores. Music taste may contribute a small heuristic adjustment internally; it does not produce a reported personality assessment.

Quality notices flag low differentiation, close leading scores or response patterns. These rules are not validated lie scales. Do not overinterpret small ranking differences. A score of 80 is neither an 80% chance of success nor a population rank.

Programmes and AI: catalogue data and readings of public programme pages form the basis. With sufficient content evidence, the optional personal AI comparison can move up to 30 points from the computed score and provide reasons. Without that basis, a catalogue-only explanation must not alter ranking. AI responses may vary or be wrong; check course content and admissions with the institution.

Data flows and deletion

This overview describes application behaviour. “Processed locally” does not mean that every feature keeps all information on the device. See the privacy notice for controller information, measurement services and rights. Privacy →

Security and institutional review

Implemented in code: separate consent before personal AI requests, a server-side consent-field check, unlock checks on paid endpoints, bounded and validated AI payloads, HTTPS calls to Infomaniak, and no HTTP caching of personal AI endpoint responses. This is not an independent security certificate or a guarantee about provider logs.

Still to document before institutional use: production hosting and Redis regions, HTTPS configuration, access controls, actual log and backup retention, subprocessors, contractual and deletion commitments from every provider, and any required impact assessment and internal approvals. This documentation does not replace that review. Server location alone does not answer these questions.

Until internally approved, demonstrate only with fictitious cases. The questionnaire can be used without personal exports or personal AI requests; that workflow is not automatically approved under an authority’s rules. Do not send client data to AI where internal rules prohibit it.

Introduction for advisers

Self-study introduction, not certified training. Objectives: explain results, recognise limitations and demonstrate data flows. Personal training or professional support may be requested; no accredited training programme is claimed.

1. Read the manual and complete a questionnaire using fictitious responses. Explain the interest profile, ranking and personality means separately.
2. Case exercise: two fields score 78 and 80 and the person likes both. Do not infer an aptitude decision; compare curricula, trial visits and unanswered questions. Two points are not an established meaningful difference.
3. Privacy exercise: explain what remains local without uploads, what programme search sends, and what AI consent adds. Demonstrate local deletion and explain why shared links and provider logs do not disappear with it.
4. Knowledge check: is 4.50/5 the 90th percentile? No. Does the RIASEC model establish this ranking’s validity? No. Does removing a name automatically make an AI payload anonymous? No. Seek professional support before using a real counselling case if unsure.

Contact for professional enquiries: team@whatshouldistudy.com

Empirical evaluation plan

Proposed, not completed. Publishing this plan does not collect research data.

1. Specify the protocol first: involve an independent psychometric expert, define target population and use, and freeze software, item and language versions. Preregister hypotheses, analyses, exclusions, missing-data handling and success criteria. Derive sample size from precision or power analysis, not from favourable results afterwards.
2. Review items and translations: cognitive interviews with the target population, expert content review and translation review. Document unchanged and adapted items separately.
3. Reliability and structure: examine scale consistency with uncertainty intervals, test–retest stability at a prespecified interval and factor structure. Test questionnaire, source analysis and AI separately and together; record AI model and prompt versions.
4. Validity and added value: compare with suitable established interest and personality measures; plan independent adviser assessments and later study satisfaction or progression as separate criteria. Test whether digital sources and AI add value beyond the questionnaire. Do not tune weights on evaluation data.
5. Norms and fairness: define a relevant German-speaking target population and recruitment strategy for any future norming. Report age, educational pathway, region, language, collection year, selection bias and sample counts; examine relevant groups and measurement invariance. Do not label a volunteer online sample representative.
6. Before recruitment: establish separate research consent, data protection arrangements, retention, rights and any required ethical review. Publish negative findings, limitations, conflicts of interest and versions too. Seek a new institutional assessment only with verifiable evidence.

Items, references and version

Revision 1.0, 14 October 2026: first consolidated professional documentation; unsupported personality norms removed; complete personality responses required; AI consent checked at the endpoint. Earlier result links without response means are not converted into new norm scores. Model changes require renewed review and documentation.

• Mini-IPIP scoring key · Donnellan, Oswald, Baird & Lucas (2006)
• IPIP: scoring positive and negative items
• IPIP: norms and datasets
• Item definitions and scale scoring · Ranking model · Programme fit