AI evidence extraction

Darwin Education: Architecture-First Adaptive Learning With Psychometric and Safety Governance

Demetrios Chiuratto Agourakis, Isadora Casagrande Amalcaburio · 2026

AI-generated evidence extraction, verified across multiple analytical personas. Not a substitute for the peer-reviewed original.

5/10

Relevance

1/4

Quality (LMQS)

Evidence

Citations

0.00

FWCI

This is an AI analysis. Read the peer-reviewed original at the publisher: https://doi.org/10.36227/techrxiv.177155954.44089129/v1

Methodology & findings

Study design

Production system implementation and documentation with architecture-first design prioritizing psychometric grounding and safety governance.

Primary method

Design science / Architecture-first approach with emphasis on psychometric grounding and safety governance

Main result

The system implements "four subsystems: (1) IRT-based psychometric inference from ENAMED (Brazilian Medical Education Assessment) microdata, (2) learning-gap detection via multidimensional difficulty modeling, (3) LLM-driven adaptive question generation with citation verification, and (4) multi-gate validation pipelines with mandatory human review for edge cases." Runtime corpus enumeration reports "215 unique diseases and 602 unique medications (14.68% and 32.28% duplicate fractions, respectively)" from Darwin-MFC submodule exports.

Research paradigm

Design Science / Engineering

Author conclusions

The authors conclude that their approach represents a paradigm shift in tutoring system evaluation: AI-based tutoring systems are "typically evaluated at the interface layer (generation quality, user engagement) rather than the control layer (measurement rigor, governance transparency, safety instrumentation)." The Darwin Education system "prioritizes psychometric grounding and safety governance over generative fluency" and demonstrates that "All code, data, and reproducibility artifacts are openly available (GitHub, Zenodo DOI)," emphasizing transparency and reproducibility as core values.

Risk of bias

No user study conducted; no randomization or control group; no educational efficacy claims tested; single production implementation without comparative evaluation; potential selection bias in ENAMED microdata composition not addressed.

Limitations

The authors explicitly state: "This implementation report makes no educational efficacy claim." This represents a significant limitation in that the paper documents a system architecture and safety framework without providing empirical evidence of learning outcomes or educational effectiveness.

Open questions raised

The paper identifies a gap in how AI-based tutoring systems are evaluated, noting the lack of emphasis on measurement rigor, governance transparency, and safety instrumentation at the control layer. Future directions would involve educational efficacy studies, which are explicitly not addressed in this implementation report.

Data: not_statedCode: not_statedExtracted from: pdfAgreement 70%

Explore related topics

Responsible AI Use Scholarly Infrastructure Design Science