SKILLS-XR — Pedagogical AI interventions to reverse declines in maths, reading, and science HORIZON-CL2-2027-01-TRANSFO-06

SKILLS-XR will (i) identify structural causal mechanisms behind declining mathematics, reading, and science skills and (ii) generate policy-grade causal evidence on effective, scalable, and costed interventions—leveraging our demonstrated expertise in immersive learning, AI-supported personalised instruction, metaverse/digital escape-room microlearning, and engagement-oriented formats (e.g., pedagogical memes, inquiry scenarios).

3) Research questions (aligned to the call scope)

RQ1 (Structural diagnosis): Which combinations of changes in teaching methods, curriculum/assessment practices, teacher shortages/conditions, investment levels, and school climate best explain the cross-country decline and widening gaps?
RQ2 (Digital shift): What is the causal role of digital distractions and the shift from paper to screens in reading comprehension and sustained attention?
RQ3 (Embodied cognition): What measurable role do handwriting and drawing play in conceptual development in maths/science and deep reading?
RQ4 (Skills interaction): How do self-regulation, socio-emotional skills, learning skills, and school climate mediate achievement trajectories?
RQ5 (Heterogeneity): How does increasing classroom heterogeneity (incl. migration-related) affect teaching organisation and outcomes, and which supports reduce friction while improving achievement?
RQ6 (Institutions & families): What roles do parental involvement (levels/types) and cultural institutions (libraries/museums) play, and at what cost?

4) Approach and methods

A. Multi-source causal diagnosis (“why did skills fall?”)

• Integrate PISA microdata with national assessments, teacher workforce indicators, school environment data, and policy timelines into a harmonised EU causal evidence base.

• Use quasi-experimental designs around reforms and shocks (e.g., staggered curriculum/assessment changes; device policies; teacher supply shocks): difference-in-differences, synthetic control, regression discontinuity where applicable; plus causal ML for heterogeneous effects (e.g., SES × gender intersections).

• Complement with youth-centred qualitative work (focus groups, diaries on reading/attention habits) to validate mechanisms and interpretability.

B. Intervention portfolio (“what works, for whom, at what cost?”)
We will evaluate a small number of high-leverage, scalable interventions grounded in our competence:

1. Digital Escape-Room Microlearning for maths/science reasoning and reading comprehension (adaptive puzzle paths).

2. AI-supported inquiry tutoring (prompted questioning, feedback, self-regulation nudges; teacher-in-the-loop).

3. Low-tech cognitive anchors: structured handwriting/drawing routines embedded into maths/science and reading lessons (counterbalance screen-based fragmentation).
   Designs: cluster RCTs (schools/classes), plus pragmatic trials in real conditions; accessibility ensured via UDL-informed design and multimodal content.

5) Work plan (WPs) and deliverables

• WP1 Governance & data access (GDPR-by-design, data-owner agreements, youth advisory board).
  D1: Data management + ethics package; stakeholder cooperation MoUs.

• WP2 Structural causes modelling (cross-country causal analysis; mechanism briefs).
  D2: “Structural Causes Atlas” (low/top achievers; SES×gender).

• WP3 Co-design of interventions (teachers + students; accessibility; implementation supports).
  D3: Intervention protocols + teacher PD modules + preregistrations.

• WP4 Impact evaluation (RCT/quasi-experiments; heterogeneous effects; robustness).
  D4: Impact report + replication package (code, templates).

• WP5 Cost-effectiveness & policy translation
  D5: Policy Playbook + cost calculator + guidance for future evaluations.

• WP6 Clustering, dissemination, uptake
  D6: Policy labs, evidence briefs, practitioner toolkits; cross-project meta-analysis contributions.