phea-atlas: a calibrated probabilistic decision atlas for 5-component high-entropy alloys (1,086,008 systems)

Probabilistic decision-support atlas accompanying the manuscript "A Physics-GuidedBayesian Framework for Decision Support in High-Entropy Alloy Discovery"(Knowledge-Based Systems). It assigns a calibrated P(HEA) ∈ [0, 1] to every5-component alloy that can be drawn from a 44-element pool — C(44,5) = 1,086,008subsets in total. P(HEA) is produced by a three-layer Bayesian framework: a Miedema semi-empiricalprior → a Gaussian-process posterior trained on experimental binary mixingenthalpies (σ_ΔH propagated exactly through the full GP covariance) → a calibrationlayer (logistic regression + temperature scaling, ECE = 0.041) trained on n = 433experimental HEA outcomes. Key facts: 1,086,008 subsets; 1,077,940 with ≥1 valid composition; 8,068 all-out-of-range. Distribution of mean_P over valid subsets: [0, 0.1) 95.9%, [0.1, 0.3) 4.1%, [0.3, 0.5) 99 subsets, >0.5 none. Top subset: Co-Cr-Cu-Fe-Ni, mean_P = 0.464, max_P = eq_P = 0.935. Reproduction code: phea-atlasFull column dictionary and integrity checksums: see ATLAS_README.md (file preview).