Hybrid AI Models for Materials Discovery with Versioned Laboratory Datasets: An Audit-Ready Methods Framework

Abstract

Materials-screening datasets often mix experimental batches, computed descriptors and inconsistent processing metadata. An evaluation that overlooks these constraints may reach conclusions that cannot be reproduced. The research question is whether it is possible to examine whether hybrid prediction models remain reliable when material families and laboratory batches are held out. Previous work provides context for sampling, input validation and study boundaries [3–7]. The planned analysis uses a reproducibility-focused benchmark built from versioned materials datasets and documented processing histories. Prior work on storage operations provides technical context for validation [1], while complementary systems evidence motivates service-continuity checks [2]. Under matched evaluation conditions, the protocol compares physical-descriptor baselines, purely learned models and hybrid models combining domain constraints with machine learning. Methodological sources are used to specify transparent comparison and reporting practices [8–12]. The audit-focused extension preserves processing logs, model decisions and exception handling for independent review. Performance will be judged using out-of-family prediction error, calibration and stability of candidate rankings under data revisions. Interpretation will account for recipe duplicates, provenance gaps, leakage across related materials and sensitivity to batch-specific artefacts. Uncertain outcomes would be adjudicated with a predefined rubric, rather than silently recoded as successes. The supplementary technical literature is used to identify plausible failure mechanisms [13–17]. No study result is assumed in advance; the analysis is designed to show both when the method helps and when it fails, with reporting context from the remaining sources [18–23].