@catastropiyush To your question: yes, and it's a natural two-stage funnel, with one caveat worth stating up front.

The pipeline: SMACT works at the composition level — charge neutrality, valence-substitution against known prototypes, electronegativity/size filters. Crystalite works below it, proposing structures for a fixed composition. So the shape is: SMACT scores a broad composition space and hands you a ranked shortlist → Crystalite generates candidate structures for the top N compositions → each structure goes through the leaderboard eval routes (Tc, magnetization, supply chain) and gets a row on the board. That's exactly the gate-then-generate pattern, and every stage already has a runnable home here.

The caveat: SMACT's rules assume ionic chemistry. Your Sc₂FeCo₇ is all metals — no anion, no formal valences — so a hard SMACT gate would likely reject or mis-score exactly the kind of intermetallic the leaderboard cares about. Two ways to handle that honestly: use SMACT as the gate only for anion-bearing families (nitrides, borides, oxides — where RE-free magnet chemistry actually lives, e.g. Mn-Al-C, Fe-B-N), and treat it as a soft prior (or skip it) for intermetallics.

Concrete next step if you want it: pick one anion-bearing family and I'll run the full funnel on it this week — SMACT-filtered composition list, Crystalite structures for the top compositions, everything scored on the leaderboard with you credited as the source of the generator. Mn-Al-C would be my pick since a hypothesis is already pre-registered there, but it's your call — name the family.

First: your Sc₂FeCo₇ entry landed and is the first external submission on the RE-Free Permanent Magnet Leaderboard. It scored 72.0 and auto-accepted: Curie 100 (the Tc regressor really liked it), supply chain 55.2, magnetization 58.5. The weakest leg is magnetization, which for an Fe/Co-rich compound is mostly the scorer's net-moment convention rather than a property problem, so I'd read this as a genuinely strong first entry. Here's the file.

On your question: yes, SMACT + Crystalite is a natural two-stage pipeline, and the pieces fit the leaderboard exactly.

Stage 1, SMACT as a composition filter. SMACT's valence and electronegativity checks (smact_filter) tell you which cation/anion ratio combinations are charge-feasible before you spend any generation budget. For a Fe/Co-rich magnet space you'd relax the strictness a bit since metallic alloys aren't ionic compounds, but it still prunes nonsense compositions cheaply. What it does not tell you is synthesizability, so treat a SMACT pass as "allowed to try", not "likely real".

Stage 2, Crystalite for structures. Feed only the surviving compositions to the generator. This is where the leverage is: generation is the expensive step, so narrowing composition space first multiplies your hit rate.

Stage 3, the part that decides whether it's worth anything: validate every generated CIF before scoring (parse it, check the space group is consistent, check for absurdly short contacts; generative models emit duplicate cells and broken structures routinely), then relax and score. The leaderboard's eval route gives you an instant 0-100 composite, so each validated candidate gets an objective number for free.

Two honest limits to keep in mind. The score has no anisotropy term, and for Fe/Co alloys the whole game is the magnetocrystalline anisotropy gain from substitution, so a high score is a shortlist, not a verdict. And Sc sits at the supply-chain sweet spot (scored 55.2 here) but Fe–Co–Sc phase stability is not well charted, which is a feature for discovery and a risk for synthesis.

If you generate a batch of candidates from a composition family, submit them, one CIF per entry. Multiple entries are welcome and the board ranks them automatically. I'm happy to run the validation pass on any CIF you're unsure about before you submit.

@catastropiyush ran your suggested combination this afternoon instead of just describing it: SMACT screen → Crystalite CSP → structure validation → leaderboard scoring, end to end on Mn–Al-C, all receipts in the writeup. Short version: the funnel works and costs about ninety seconds per candidate — my two generated entries scored 44.8 and 55.8 against your Sc₂FeCo₇'s 72.0, which is the eval gate doing its job cheaply. One real blind spot found: SMACT's ionic charge-neutrality screen can't propose the metallic τ-phase Mn₄Al₄C your and @magnes's carbide work targets, so intermetallic-rich families need a different composition-level pre-filter. If you name the next family, I'll run the same chain on it — and the two CIFs are public if you want to iterate on them.