Trigger: AI may influence an adult learner's pathway, credential, or skills record.
Skills and adult learning AI · supporting context
Learning AI is assessed
under the context it actually touches.
A supporting context, not a lead one. Where an adult learner's pathway is at stake, this is the evidence set. Where a learner is under 18, the controlling regime is minor-facing AI — GDPR Art. 8, DSA Art. 28, and the Art. 5 prohibition on emotion inference in education.
Sensitive context
Where learning AI sits — and where it defers.
Adult skills and learning AI is a supporting context in this platform, and it is treated as one. It matters where a recommender narrows an adult's pathway, where an adaptive assessment produces a credential that can be appealed, or where a RAG assistant grounds guidance on stale course content. It stops mattering as a category the moment the learner is under 18: at that point the controlling instruments are GDPR Art. 8, DSA Art. 28, and the AI Act Art. 5(1)(f) prohibition on emotion inference in education, and the system belongs on the minor-facing AI page — not this one. Use this context for workforce learning, apprenticeships, and adult skills. Route anything a child can reach to the minors regime, which is stricter by design.
Data categories in scope
People affected
Risk scenarios
What typically goes wrong.
Specific failure modes seen in this sensitive context — without structured evidence.
A skills graph recommender suggests pathways with measurable demographic disparities.
No Group Disparity Analysis evidence records. No subgroup breakdown. Disparate impact undocumented. Procurement cannot evidence fairness review.
An education RAG assistant answers questions using outdated course content.
No Grounding and Response Quality evidence records. Document boundary not configured. Learners receive incorrect course guidance.
A workforce learning AI platform sends learner progress data to a third-party recommendation engine.
No DPA confirmed with recommendation API provider. Learner data leaves platform without structured evidence of legal basis.
An adaptive assessment AI produces results that cannot be reproduced for a learner appeal.
No Model Registry Trace active. No lineage. Past assessment context irretrievable. Appeals process undermined.
Border control, applied here
Scan the system. Issue the passport. Grant the visa.
What the scan finds in this context, what the passport records, and what conditions the visa attaches.
Currents
the scanWhat AI is running, and what do we not know about it?
Currents opens the file: which learning systems influence an outcome an adult can appeal, which recommenders narrow a pathway, which RAG assistants ground on stale course content — and which of them are reachable by a learner under 18, in which case they leave this context and enter the minors regime.
Ripples
the passportWhat is this system, and what is proven about it?
The Ripple records the learner data boundary, subgroup performance where the system influences an outcome, retrieval grounding for the assistant, and lineage sufficient to reconstruct an assessment decision at appeal. Learning efficacy is not among the things it records, because it is not among the things we measure.
Droplets
the visaMay it operate HERE, on THIS data, under WHAT conditions?
The Droplet clears the system for one cohort and one purpose: adult learners only, human assessor review before any credential or formal outcome is recorded, grounding above threshold, and an expiry that forces the fairness evidence to be refreshed each cycle.
Scope
What needs a Ripple.
Stakeholder workflow
From trigger to access decision.
Scan
Art. 6 route · Annex III test
Evidence call
Annex IV · Art. 10 bias file
Seal Review
DPO · CISO · Art. 14 owner
Droplet
Conditions bound · expiring
Re-scan
On retrain, drift or model swap
Scan
Art. 6 route · Annex III test
Evidence call
Annex IV · Art. 10 bias file
Seal Review
DPO · CISO · Art. 14 owner
Droplet
Conditions bound · expiring
Re-scan
On retrain, drift or model swap
DPO
“Learner performance and progress data will be processed by an AI recommendation engine.”
Confirm legal basis, DPA, and data minimisation evidence. Review before clearance.
Head of Learning & Development
“A skills matching AI is being procured for workforce development.”
Require Ripple with fairness signals and human oversight conditions for high-stakes pathway decisions.
Procurement
“A skills or learning platform claims AI-powered personalisation.”
Request structured Ripple. Require RAG grounding evidence and data boundary confirmation.
Access decisions
Context Droplet conditions.
The access decisions that apply in this sensitive context — and the evidence conditions that produce them.
- Adult learners only — under-18 reach routes to the minors regime
- Fairness evidence provided for outcome-influencing features
- RAG grounding evidence present
- Human assessor review gate for high-stakes recommendations
- Fairness evidence missing for outcome AI
- RAG grounding below threshold
- Learner data leaving platform without DPA confirmation
- AI influences formal qualifications, credentials, or grading
- A human assessor must verify before a formal outcome is recorded
- Raw learner data exported without legal basis
- No explainability for assessment outcomes subject to appeal
Measurement
Evidence families we can structure.
The measurable evidence categories relevant to this context and the evidence signals they produce.
Data Quality
Completeness, consistency, and freshness of training and inference datasets for skills and learning AI.
Fairness & Subgroup
Subgroup performance and disparate impact for recommendation and outcome-influencing AI.
RAG Grounding
Faithfulness, relevance, and contextual precision of education RAG assistants — without exporting raw course content.
Privacy
Legal basis for learner data processing, DPA status, and data boundary evidence.
Lineage
Dataset and model lineage enabling reconstruction of recommendation or assessment outputs for appeals.
Human Oversight
Evidence that educators or assessors review AI outputs before formal educational outcomes are recorded.
Honest scope
What remains not assessable.
AffectLog does not overclaim. These items require external expertise, regulatory process, or long-term study.
Learning efficacy or educational effectiveness
Whether an AI learning tool improves outcomes requires a controlled evaluation study with appropriate methodology.
Instead: Commission an independent learning efficacy evaluation before deployment at scale.
Credential or qualification recognition
Credential recognition requires awarding body or regulatory authority processes — not an evidence platform.
Instead: Engage the relevant awarding body or regulatory authority for formal recognition processes.
Example
Sample Ripple for this context.
SkillsMap Recommender v3
Workforce skills pathway recommendation · L&D Platform
Access conditions
What we will not overclaim
AffectLog provides technical and operational evidence for education and skills AI. We do not claim learning efficacy, educational effectiveness, or credential recognition. We show data boundaries, fairness signals, and what review conditions are required.
Common questions
Questions this context raises.
“Our skills platform uses AI responsibly — we have an ethics policy.”
An ethics policy is a commitment document. AffectLog structures measurable evidence: fairness evidence records, RAG grounding scores, data boundary configuration, and lineage — so DPO and procurement can verify rather than rely on policy statements.
“We don't collect sensitive learner data — only course completion rates.”
Course completion data combined with skills profiles can still reveal demographic patterns and influence opportunity. Even low-sensitivity data categories benefit from structured evidence when AI is used to make recommendations.
Get started
Map every AI tool shaping
learner pathways and skills.
Identify which education and skills AI systems need evidence, which RAG assistants need grounding tests, and which recommendation engines need fairness signals before they influence the next cohort.
AffectLog provides technical and operational evidence. Not learning efficacy claims, educational certification, or regulatory approval.