Data and code for: Quantifying the propagation of default-setting errors in machine-learning screening pipelines for halide perovskites

Analysis code and figure-generation scripts underlying the seven quantified default-setting errors reported in the accompanying manuscript. Each measurement is a difference between two protocol variants evaluated under otherwise identical settings.</p><p>Contents: the corrected octahedral-connectivity filter and its validation against reference structures of known connectivity; first-principles workflow scripts (GPAW) for the tilt scans, equation-of-state fits, c/a scans, full-zone phonon calculation and symmetry-breaking relaxations; the 209 generated candidates with predicted properties; the machine-learned-potential tilt screen and its DFT calibration; and all raw result files from which the reported numbers are computed.