Scandium-Dataset Roadmap

Vision: The field's reference open dataset for AI-driven solid-state battery materials discovery — combining scale (266k+ entries) with purpose-built SSE screening properties that no existing open dataset provides.

Strategic plan: See docs/ssb_strategic_roadmap.md for the full literature-backed strategy document with ~20 references.

v0.1.0 — Foundation Release (Done)

Focus: Clean, validated, audited general inorganic materials aggregation

Aggregate MP, OQMD, JARVIS-DFT (266,732 entries)
Coordinate repair (171,780 OQMD entries)
Symmetry pass (171,764 space groups)
Volume/density repair (73,480 entries)
11-gate validation pipeline
Multi-source deduplication
Quality scoring with 5 sub-scores
Three-tier system (Gold / Validated / Raw)
Battery (82,925) and electrolyte (41,665) subsets
Benchmark splits (4 types)
Complete 8-phase audit
Reproducibility scripts
Full documentation (12 audit reports, trust report, dataset card, known issues)
Docker support
Test infrastructure
Licensing resolved (per-entry license field, LICENSE_BREAKDOWN.md)
SSE proxy features (carrier_fraction, volume_per_carrier, fe_per_carrier, electronic_insulation)
SSE family classification on all entries (garnet/NASICON/argyrodite/LGPS-type/etc.)
Electrochemical window computation script
Baseline benchmarks (RF+Ridge, 8 experiments)
Related work analysis, paper outline, SSE readiness assessment
v0.1.0-rc.2 release to GitHub + HuggingFace

Phase 1 — Cheap, High-Leverage, No New DFT (Complete ✓)

Focus: Add structural descriptors, stability windows, SSE screening schema → milestone v0.2.0

Run CAVD channel-dimensionality analysis — schema exists, no data (0/190k entries)
Build electrochemical stability windows — 1,814/266k entries have data (partial)
Compute free-volume, packing fraction, coordination descriptors — status unknown
Create ssb_screening schema block across all entries (266,732 have schema, all fields null/empty)
Expand garnet family enrichment (target 500+) — still at 41, no expansion done
Publish sse_candidate_score based on gates 1–2 (thermo + electronic) — 2,354 entries scored (10-19 range)
Energy_above_hull for JARVIS entries via internal convex hull — 0/25,673 entries computed
Oxidation state prediction for all entries (BVA + heuristic) — status unknown
Commercial-safe edition extraction (MP + JARVIS only, ~95k entries) — 94,952 entries confirmed
Mechanical property proxies — schema exists, 0 entries have bulk/shear moduli
Formation energy corrections for OQMD entries

Phase 2 — MLIP-Driven Scale-Up

Focus: BVSE + MLIP-NEB migration barriers + elastic tensors → milestone v0.3.0

Implement BVSE migration barrier proxy — rewritten to use bvlain (softBV); passes 5/7 sanity cases, worst error 0.12 eV
Setup script for CHGNet/MACE-MP-0/M3GNet infrastructure (scaffold written, not yet run)
Extend frozen splits to conductivity/stability tasks (generated from BVSE data, valid)
Run MLIP-NEB migration barriers on CAVD+BVSE-filtered subset — scaffold only, not executed
Run MLIP-based elastic tensor estimation — not started
Publish first full sse_candidate_score (all 5 gates) — gates 3-5 only partially populated
Leaderboard with baseline model results
Full BVSE run on all 107k Li/Na entries (requires ~3 CPU-days)

Phase 3 — Gold-Standard Validation Layer

Focus: DFT verification + experimental ground truth → milestone v0.4.0

Ingestion scaffold for experimental conductivity data (script written, needs correct URLs + actual fetch)
Stratified DFT-NEB/AIMD verification sample (50–100 entries)
Full DFT elastic tensor calculations for Strict-Gold validation subset
Ingest OBELiX (Therrien et al. 2025, NRC-Mila) as experimental gold subset
Ingest Hargreaves et al. 2023 npj Comp. Mater. conductivity database (DOI: 10.1038/s41524-022-00951-z)
Publish per-family, per-method accuracy tables
Cross-reference with experimental ICSD entries
Compute phonon properties for stability validation

Phase 4 — Community & Publishing

Focus: DOIs, challenge, community adoption → milestone v1.0.0

Dataset DOI with versioned releases (Zenodo)
Publish composite sse_candidate_score formula openly
Community submission workflow for new entries/properties
Web API for dataset querying
Automated monthly source update pipeline
Integration with materials discovery workflows
Community governance model
Target journal submission (Scientific Data or NeurIPS D&B)