A synthetic GAGG-like waveform dataset for amplitude-dominated gamma-source classification

Each event contains 2,000 float32 waveform samples, a source-category label and a synthetic energy annotation in MeV.The nominal record window is 400 ns at 5 GHz. The endpoint-inclusive generation grid has an actual spacing of 400/1999 ns.

All classes share one nominal decay shape, with source-dependent energy distributions, Poisson amplitude fluctuations and amplitude-dependent Gaussian noise. The stored records have truncated leading portions and exact trailing-zero padding. They are analytic synthetic data, not experimental measurements or particle-transport simulation output.The supporting archive includes a data dictionary, generation protocol, fixed training/validation/test indices, aligned event metadata, quality-control reports, analysis code, fitted scalar classifiers, predictions and figure source data.