The compositional space for refractory HEAs is vast and largely unmapped — outside the Cantor alloy and its close derivatives, little is experimentally anchored. The HEA tool is therefore built as a compositional screener, not a property predictor: its job is to eliminate the unviable and rank the plausible within an intractable space, flagging extrapolation honestly rather than claiming confidence it can’t yet support. The case study below shows that screening role in context.

Case study #3 (Cantor-scope): The HEA tool screens a candidate composition space — variations in a chosen alloy family — using two working components today: analytic phase-stability rules (VEC, δ, Ω) as a pass/fail gate, and Bayesian multi-objective optimization with a physics-informed prior to identify the Pareto front across competing objectives like strength and oxidation resistance. Every shortlisted composition carries a Mahalanobis-distance flag showing whether it falls inside or outside the compositional domain the model has actually seen data for.

Validated scope right now is the Cantor family (CrMnFeCoNi and its documented derivatives) — the most experimentally characterized HEA system in the literature, which makes it possible to check the pipeline’s rankings against real held-back data rather than assume they’re correct. The plan: source real published compositions, hold a subset back from fitting entirely, run the pipeline blind, and report directional ranking consistency against known results — not absolute property predictions.

Two things this tool does not yet do: CALPHAD-based phase screening (not implemented — analytic rules substitute for now), and anything outside Cantor-adjacent composition space. Refractory and nuclear-relevant systems are a different, harder anchoring problem with far less experimental data available, and any output there would carry an extrapolation flag by construction — meaning low confidence, honestly labeled, not a ranked recommendation ready for fabrication decisions.