Companion data for "Predicting the Neutrino Mass Ordering Using Neural Networks": pre-trained models and training dataset

Companion data for the study comparing a neural-network classifier against standard log-likelihood (LLH) minimisation for determining the neutrino mass ordering (Normal vs Inverted) in a NOvA-like long-baseline experiment, described in https://arxiv.org/abs/2606.03745This archive (companion_data.tar.gz) contains the pre-trained Keras models and the 2-million-sample training dataset needed to reproduce the figures in the paper without regenerating the data from scratch.Unpack it into the root of the analysis code repository. Contents - train_sample.pkl — 3-flavour training dataset (2 M Poisson-fluctuated pseudo-spectra) - MO_model.h5 — trained 3-flavour mass-ordering NN classifier - dm2_model.h5 — trained 2-flavour Δm² NN regressor - NO_output.pkl, IO_output.pkl — 3-flavour 1000-toy results - 2f_results.pkl — 2-flavour 1000-toy + bootstrap resultsAnalysis code: https://github.com/universityofsussex-mps/nn-vs-llh-mass-ordering