An ensemble of open binary CALPHAD assessments quantifies the uncertainty of the Nd2Fe14B feasibility threshold and surfaces inconsistencies invisible to single-assessment users

A single merged ternary description built from open binary CALPHAD assessments is only one of many that can be constructed from the published literature. For Nd–Fe–B, three Fe– Nd, seven Fe–B and four Nd–B assessments are freely available, giving 84 candidate ternary descriptions. We screen all 84 for feasibility of Nd2Fe14B at the stoichiometric composition over 800–1600 K and report the distribution of feasibility thresholds. Seventy-eight of the 84 combinations (92.9 %) are mutually compatible under a strict (1 J mol−1) merge-then-edge-check protocol. The six failures all involve the Witusiewicz 2019 Fe–B assessment—the only one that treats boron as purely interstitial in the disordered bcc, fcc and hcp phases—and arise because the merge silently alters the sublattice model of those phases; each fails by only 6.2 J mol−1 at a single test point. Re-admitting them leaves the range unchanged and shifts no median by more than 0.07 kJ. At 800 K the ensemble threshold spans−9.26 to−8.33 kJ/mol (per mole of atoms; median−8.79, spread 0.94). The two available full ternary assessments (Hallemans 1995 and Chen 2019) both fall inside this range, with Hallemans coinciding with the median to within 0.01 kJ. The single-value threshold of−9.2 kJ/mol used in our prior Stage 3 work sits at the more-negative extreme of the ensemble. The spread grows from 0.94 kJ at 800 K to 2.35 kJ at 1600 K. We argue that the ensemble is a minimal-assumption description of the current open thermodynamic data for Nd–Fe–B, and that single-value CALPHAD claims should be reported together with the ensemble spread around them