IRT / 2PL / optional 3PL model. 2PL: P(Y_ui = 1) = sigmoid(a_i * (theta_u - b_i)) 3PL: P(Y_ui = 1) = c_i + (1 - c_i) *

Small LLM wrapper used by the Action module. The default provider is ``mock`` so the repository can be smoke-tested without network access or

Encode logs in the same form used by the original DNeuralCDM code.

Action module: generate and parse four learner-simulation tasks.

DNeuralCDM-produced dynamic knowledge proficiency loader. Expected JSON shape is compatible with the old script: ``{student_id: [[[k_0, k_1,

Memory module aligned with the paper pipeline. It contains factual memory, short-term memory, long-term memory, KCG-based memory reinforcemen

Learner Profile module. It consumes the profile file produced by ``Code/prepare/build_profile.py``. The six fields follow the paper implement

Knowledge Concept Graph (KCG) wrapper. The paper uses the KCG/RCD-derived relation graph to judge whether two concepts are related when reinf

data loader for training