Reproducibility dataset for a PRISMA 2020 systematic review of machine-learning approaches to self-healing cloud infrastructure (autonomous detection, diagnosis, decision, and recovery), covering 341 included studies published 2022 to 2026 and 139 excluded records. The archive contains the master coded dataset (canonical_studies.csv), per-paper extraction records with CASP and PROBAST-adapted risk-of-bias scoring rationale, derived cross-tabulations, the title and abstract screening decision trail, a statistical re-derivation script for the Fisher exact and Wilson 95 percent confidence interval gap comparisons, and figure-generation source. Data and code only; the full methodology and rubrics are described in the associated article and its supplementary appendices.