qMR-FailureBench: A Benchmark for Explainable Failure Forecasting and Counterfactual Correction in Quantitative MRI

qMR-FailureBench provides standardized datasets, trained models, evaluation scripts, and baseline results for testing uncertainty quantification and failure detection methods in quantitative MRI under entangled physical corruptions. Contents:- MRF benchmark (9,990 test samples, 1000 timepoints)- MRS benchmark (2,000 test samples, 2048 points)- 12 pre-trained model checkpoints- 13 paper figures and 9 metric JSONs- Standard evaluation script (evaluate.py)- Complete model source code Evaluation tasks:1. Failure Detection (AUROC)2. Corruption Attribution (F1)3. Severity Estimation (MAE)4. Counterfactual Repair (ΔMAE)5. Sim-to-Real Calibration (Spearman ρ) Corruption types: B₀ off-resonance, B₁+ transmit scaling, k-space motion (entangled).