Compliance · EU AI Act Art. 13
AI Literacy Index — Methodology Note
How Keypra computes ALI scores and tier labels.
Core declaration
The ALI score and tier labels are computed by a deterministic weighted formula in a PostgreSQL database function. No AI model is involved in computing the ALI score. The formula is fully auditable.
What the ALI Score Is — and Is Not
| The ALI score IS | The ALI score is NOT |
|---|---|
| A weighted index of observed learning activity | A measure of general intelligence or cognitive ability |
| Evidence of engagement with Keypra's curriculum | A validated psychometric assessment |
| A compliance evidence trail for Art. 4 AI literacy obligations | A performance review or employment appraisal instrument |
| Fully reproducible from raw activity records | An AI-generated judgment |
Important for deployers: The ALI score documents that learning activity occurred and to what extent. It does not certify competency. Employment decisions must not be based solely on this score without independent human review.
The formula
ALI = (0.30 × Foundation)
+ (0.25 × Retention)
+ (0.20 × Practice)
+ (0.15 × Awareness)
+ (0.10 × Recency)Each component is independently capped at 100. Final score rounded to 1 decimal place.
Component definitions
Foundation (30%)
- Formula:
MIN(100, (lessons_completed / total_lessons) × 100) - Source:
lesson_progresswherestatus = 'completed' - Range: 0–100
Retention (25%)
- Formula:
AVG(score)for completed retention checks - Source:
learner_retention_checks - Range: 0–100; null if no checks taken
Practice (20%)
- Formula:
MIN(100, (quests_in_last_90d / 12) × 100) - Source:
weekly_quest_progresswherecompleted_at >= now() - 90 days - Range: 0–100
Awareness (15%)
- Formula:
((overall - 1.0) / 3.0) × 100 - Source:
fluency_assessments, most recent per user, scale 1.0–4.0 → 0–100 - Range: 0–100; null if no self-assessment
Recency (10%)
- Formula: step function — 100 if active ≤30 days ago, 50 if ≤90 days ago, 0 otherwise
- Source:
profiles.last_activity_at - Range: 0, 50, or 100
Tier assignment
| Tier | Condition |
|---|---|
| Insufficient evidence | No lessons, no retention checks, no quests, no self-assessment |
| Unaware | ALI < 30 |
| Emerging | 30 ≤ ALI < 60 |
| Capable | 60 ≤ ALI < 75 |
| Transformative | ALI ≥ 75 |
Tiers are assigned by fixed numeric thresholds. No AI or adaptive model is used.
Scope of AI use
| Feature | AI involved? | Notes |
|---|---|---|
| ALI score computation | No | Pure SQL weighted formula |
| Tier label assignment | No | Fixed CASE WHEN thresholds |
| Maturity Dashboard scores | No | Weighted averages of inputs |
| Mastery status transitions | No | Deterministic algorithm: score ≥ 60 → pass |
| Challenge scenario generation | Yes | AI writes the scenario text; no person is evaluated |
| Prompt response evaluation | Yes | LLM grades a user's voluntarily submitted prompt. Formative, opt-in, shown to the individual only. |
| AI Coach conversations | Yes | Conversational assistant; no measurement output |
Limitations
- Activity ≠ competency. No psychometric validation conducted.
- Self-assessment bias: Awareness component is self-reported.
- Practice decay: only quests in the last 90 days count.
- Retention nulls: members with no checks contribute 0 to weighted ALI.
- Curriculum size changes: Foundation denominator changes if curriculum grows.
- Not a legal instrument: compliance evidence trail, not a certification.
Related methodologies
- Behavioural Telemetry methodology — Art. 5(1)(f) note
- Competence scoring methodology — Art. 14 pass mark + contestability
- EU AI Act overview
- Data Processing Agreement — §8b purpose limitation
Version 1.0 — Issued 2026-04-19 — Keypra — Contact: compliance@keypra.com