Fairness & Bias Methodology

1. Scope

This methodology covers two surfaces where bias could materially affect a person:

• AI-graded competence assessments (see Competence Methodology).
• Curriculum content, persona templates, and example prompts shipped with the product.

Keypra is not a high-risk AI system under Annex III of the EU AI Act — we do not make employment, credit, education-access or public-service decisions. The commitments below are nonetheless applied as a matter of policy.

2. Demographic-blind grading

The grading rubric (4 dimensions × 0–25, pass mark 60/100) takes only the prompt text and the scenario as input. The grader is never given the learner's name, photo, nationality, gender, language proficiency, age, organisation tier, or any other protected characteristic.

3. Contestability (Art. 14 EU AI Act)

Every AI-graded result includes a contestability row prefixed [CONTEST] in the result panel. A user who disagrees with a grade can request human review at privacy@keypra.com; we acknowledge within 72 hours and complete review within one month.

4. Content review of curriculum & personas

Curriculum lessons, persona templates and starter packs are reviewed before publication for protected-characteristic stereotyping (gender, age, ethnicity, disability, religion, sexual orientation, socio-economic background). Reviews are performed by the compliance_steward role and logged against the relevant content version.

The seed library was reviewed before public launch; the review memo is archived internally and available to enterprise customers under NDA on request.

5. Annual fairness re-evaluation

Once per calendar year, Keypra runs an aggregate, anonymised score-distribution analysis to detect any systematic gap that would correlate with a protected characteristic (e.g. English-as-second-language users scoring materially lower for reasons unrelated to skill). If a gap is detected, the rubric or scenario library is adjusted and the change is recorded in this page's revision history.

6. Forbidden uses

Per the Behavioural Telemetry Methodology and Terms §10a, no Keypra score, behavioural metric or AI maturity tier may be used by an employer, lecturer or manager to make employment, promotion, discipline or remuneration decisions about a named individual. Aggregate-only on employer surfaces.

7. How to raise a concern

If you believe a Keypra output produced a biased or discriminatory result, email privacy@keypra.com with subject tag [FAIRNESS]. We will acknowledge within 72 hours and respond substantively within one month.