shamique's picture
Release v1.9.0: README.md
95fe628 verified
|
Raw
History Blame Contribute Delete
6.56 kB
metadata
license: cc-by-4.0
task_categories:
  - tabular-regression
  - tabular-classification
language:
  - en
tags:
  - materials-science
  - chemistry
  - battery
  - solid-state-electrolyte
  - ionic-conductivity
  - activation-energy
  - lithium-ion
  - DFT
  - machine-learning
  - materials-informatics
  - battery-materials
  - scandium-labs
pretty_name: >-
  Solid-State Battery Electrolyte Dataset — Ionic Conductivity & Activation
  Energy | Scandium Labs
size_categories:
  - 10K<n<100K
configs:
  - config_name: default
    data_files: data/default/*.parquet
  - config_name: verified
    data_files: data/verified/*.parquet
  - config_name: consensus
    data_files: data/consensus/*.parquet
  - config_name: gold_benchmark
    data_files: data/gold_benchmark/*.parquet

License Version Records Verified labels Pipeline Built by

Solid-State Battery Electrolyte Dataset — Ionic Conductivity & Activation Energy

The Scandium Labs Solid-State Battery Electrolyte Dataset provides literature-verified ionic conductivity and activation energy labels for solid-state battery electrolyte materials across 30,000+ Li-containing structures — a provenance-tracked, ML-ready foundation for electrolyte discovery and transport-property prediction.

A unified, provenance-tracked, ML-ready dataset of solid-state battery electrolyte materials with literature-verified ionic conductivity and activation energy labels — not just another Materials Project / JARVIS structural dump. Every one of the 183 verified transport labels traces back to its source paper, page, and evidence sentence.

Honest scope: the release contains 30838 bulk structural / thermodynamic DFT records (Materials Project, JARVIS-DFT, COD, AFLOW, NOMAD, OQMD) plus 183 human-verified experimental conductivity/Ea labels across 11 solid-state-electrolyte families, 427 materials with cross-paper consensus statistics, and a 165-record gold benchmark. The scarce verified labels are the asset; the bulk DFT rows are the structure/composition backbone for featurization.

Companion Datasets

  • Scandium Labs SSB Dataset (this dataset) — literature-verified ionic conductivity & activation energy on the electrolyte transport bottleneck.
  • Scandium-Dataset — harmonized DFT structural & thermodynamic screening foundation (267,230 materials from Materials Project / OQMD / JARVIS-DFT) for the earlier discovery-stage filter before transport prediction.

Ionic Conductivity Data

The verified config holds the literature-verified σ_RT (S/cm) and activation energy Ea (eV) labels for solid-state battery electrolytes. Each record carries its source DOI, measurement temperature, measurement method, conductivity type, and sentence-level provenance evidence.

Solid-State Electrolyte Families

All 11 major families are covered: sulfides, oxides, garnets, perovskites, NASICONs, halides, argyrodites, hydrides, borohydrides, antiperovskites, and polymer/composites.

How This Dataset Was Built

An automated multi-agent pipeline: source ingestion (8 connectors) → family classification → literature mining & LLM-assisted extraction → deterministic verification (Arrhenius consistency, unit normalization, cross-paper consensus) → human review → quality scoring → validation → release. The pipeline is fully deterministic after ingestion; no LLM calls are required to reproduce any artifact. Source code and pipeline docs: Scandium Labs on GitHub.

Data Splits

Splits are grouped by composition-family key to prevent leakage between polymorphs and doped variants of the same base composition. See splits_metadata.json in the source repository for the exact assignment.

Known Limitations

  • 98% of quality-scored records sit in a single "silver" tier — honest, not a scoring bug: experimental metadata (density, pressure, atmosphere) is sparse.
  • Antiperovskites, hydrides, and borohydrides are under-covered relative to sulfides/garnets — this reflects the field's publication volume, not a sampling choice.
  • Bulk DFT rows are not screened for solid-electrolyte relevance; check family / negative.* flags before use.
  • 150 AFLOW rows are restricted to non-commercial use (see Licensing below).

Licensing

Scandium-authored content is CC-BY-4.0. Third-party records retain their source licenses (Materials Project CC-BY-4.0, JARVIS-DFT CC0, OQMD CC-BY-4.0, COD CC0, NOMAD CC-BY-4.0, AFLOW non-commercial only). Per-record source is in identity.source_db. See LICENSE and LICENSE_BREAKDOWN.md for the full per-source table and the "AS IS" warranty disclaimer.

Citation

If you use this dataset in your research, please cite it:

@dataset{scandium_ssb_dataset_2026,
  author       = {Scandium Labs Team},
  title        = {Solid-State Battery Electrolyte Dataset — Ionic Conductivity \& Activation Energy},
  year         = 2026,
  version      = v1.9.0,
  publisher    = {Scandium Labs},
  url          = {https://huggingface.co/datasets/Scandium-Labs/solid-state-electrolyte-conductivity}
}

Links

Contact & Feedback

Open an issue on the source repository for corrections, community submissions, or questions.