Skip to content
Scientific data · ThermoML

Experimental thermodynamics, with provenance intact

thermoml-io reads, searches, and exports ThermoML without flattening away the scientific and metrological context needed to interpret each reported value.

Get started Explore the data model

ThermoPhase — Fluid and Complex Systems Research Group
Developed at
Laboratório Nacional de Computação Científica — LNCC Universidade do Estado de Santa Catarina — UDESC
Faithful ingestion Original decimal values, units, significant digits, phases, methods, and uncertainties remain explicit.
Deterministic discovery Systems and properties are ranked by stated observation counts before any result limit is applied.
Auditable provenance Source locators, checksums, retrieval metadata, citations, and recovery events travel with the data.

thermoml-io is developed by the ThermoPhase — Fluid and Complex Systems Research Group at the Laboratório Nacional de Computação Científica (LNCC) and the Universidade do Estado de Santa Catarina (UDESC).

The initial release provides:

  • safe XML parsing, official NIST JSON parsing, and optional XML validation against an explicit XSD;
  • immutable Python models for PureOrMixtureData;
  • deterministic system/property search and density-based ranking;
  • aggregation across publications without confusing document-local IDs;
  • bounded-memory analysis of local bulk .tgz archives;
  • automatic checksum-verified archive fetching and a live-updated source registry;
  • XML-first recovery through paired official JSON with explicit provenance;
  • independent monthly monitoring of bulk metadata and live Cordra IDs;
  • property-versus-condition queries ranked by publication, with one table per query;
  • descriptive rankings for systems, components, data types, properties, methods, and publications;
  • analysis-ready CSV, JSON, YAML, and Parquet export, with a separate lossless provenance view.

No experimental data is distributed

The package contains no NIST archive snapshot, publisher XML, extracted experimental table, or journal content. Network inputs and exports remain under user control and retain their original terms.

The repository notebooks download a checksum-pinned NIST snapshot only into the ignored notebooks/local-only/ directory. The snapshot is not part of the repository, documentation site, wheel, or source distribution.

Design boundary

The parsed scientific model is the source of truth. Tables are derived views, not a replacement for the hierarchical ThermoML record. Downstream numerical packages such as torch-flash should perform unit conversion and tensor construction explicitly.

Current scope

Version 0.1 decodes pure-compound and mixture datasets. Reaction datasets are detected and counted, but their typed representation is planned for a later release. This limitation is reported explicitly on each document.

How to cite

If you use thermoml-io in research, please cite its all-versions Zenodo record:

Volpatto, Diego; Marinho, Antonio; Ribeiro, Gustavo (2026). thermoml-io. Zenodo. https://doi.org/10.5281/zenodo.21825084

@software{volpatto_thermoml_io_2026,
  author = {Volpatto, Diego and Marinho, Antonio and Ribeiro, Gustavo},
  title = {thermoml-io},
  year = {2026},
  publisher = {Zenodo},
  doi = {10.5281/zenodo.21825084},
  url = {https://doi.org/10.5281/zenodo.21825084}
}

This concept DOI always refers to the project as a whole and resolves to its latest archived release. For exact reproducibility, cite the version-specific DOI shown on the corresponding Zenodo release instead.