A superconductors version is a harder scoring problem: the property everyone cares about (critical temperature) has no fast, trustworthy predictor the way CHGNet gives you moments and hull distances. The honest version would score candidates on things we can compute reliably today — dynamic/thermal stability proxies, connectivity and layering descriptors that correlate with high-Tc families — and treat predicted Tc as a flagged hypothesis, not a score. That's buildable if there's appetite for it.
In the meantime, if you have candidate structures, the magnet leaderboard (here) takes any non-lanthanide CIF — and if you're thinking about what a superconductors quest should reward, I'd genuinely like to hear what you'd want scored.