<p>Metal fatigue critically affects structural material qualification, yet fatigue datasets often remain fragmented across raw hysteresis files, spreadsheet summaries, processing records, and post-processed descriptors. This study develops a fatigue-specific relational database framework for traceable reliability analytics using an Integrated Computational Materials Engineering-compatible Process–Structure–Property–Performance (PSPP) architecture. The framework was demonstrated using an Al 6063 fatigue dataset comprising 85 sample-level records across 17 thermomechanical processing routes, supported by 159,967 cycle-level records. A Python-based ETL pipeline was implemented to standardize raw fatigue outputs, aggregate stabilized cyclic descriptors, enforce database integrity, and generate analysis-ready PSPP features. The dataset showed substantial processing-induced fatigue dispersion, with <i>N</i>_<i>f</i> ranging from 362 to 7599 cycles and CoV = 0.988. Route-family statistics revealed a clear fatigue-life hierarchy, with ECAP showing the highest mean fatigue life (6729 cycles), followed by DCT, HT, and AR conditions. Reliability analysis yielded a global Weibull shape parameter <i>β</i> = 1.223, scale parameter <i>η</i> = 2034.43 cycles, B10 life of 323.01 cycles, and bootstrap mean-life confidence interval of 1520.83-2306.27 cycles. PSPP analysis confirmed physically consistent relationships, including grain size versus log₁₀(<i>N</i>_<i>f</i>) (<i>r </i>=  − 0.788) and d^−1/2 versus yield strength (<i>r </i>= 0.781). A compact Ridge model using four PSPP descriptors achieved Leave-One-Route-Out <i>R</i>² = 0.580 and RMSE = 0.206 in log₁₀(<i>N</i>_<i>f</i>), demonstrating interpretable route-aware modelling rather than standalone deployment prediction. The proposed framework enables reproducible fatigue data governance, probabilistic reliability interpretation, and ICME-aligned database intelligence for future materials qualification workflows.</p>