T1DUJxUoM: Cohort-Fair Multimodal Digital Twins for Type 1 Diabetes

Official implementation of the Physics-Guided Training framework for Type 1 Diabetes Digital Twins. This software accompanies the research paper addressing "Autoregressive Blindness" in LSTM models. It includes: Physics-Guided Data Flooding: Algorithms to enforce causal insulin/carb dynamics. Stochastic Glucose Masking: Regularisation to prevent history-only learning. In-Silico Sensitivity Panel: A standardised validation suite (Scenarios A-D). Pre-trained Models & Demo Data: Fully reproducible artefacts included in the pretrained_models/ directory.