Version 1, machine-extracted: values were read by language-model readers constrained to quote their source and checked automatically, and the release was then audited on the printed pages by adversarial audits, an independent second review of fit-facing records and a preliminary human review of a frozen sample. Flags such as accepted and ESPEI-ready record screening results, not certification; known issues are listed in the release notes.
From open thermodynamic databases, papers read from lawful copies and pinned first-principles data, to values that each carry their source, the state they describe and the kind of evidence they are.
A CALPHAD database describes the Gibbs energy of every phase of a chemical system. Behind each database stand papers: an assessment that fitted the model, and the measurements that assessment used. ORCHID collects the open databases and reads the evidence behind them, so that a phase-equilibrium temperature, a formation enthalpy or a phase boundary can be found, compared and reused, together with the exact sentence or table row it came from.
The work runs in eight stages. Machines do the reading, but nothing a machine reads is accepted on trust: every value must be quoted, the quote must be found on the page, and fixed rules decide units, signs and the kind of evidence. The papers themselves are never redistributed.
Each card below links to the code on GitHub and to a short technical note. The full map, with a table of commands, inputs and outputs, is WORKFLOW.md.
What goes in, and under which licence. Openly licensed CALPHAD databases (the SGTE collection of 2,774 binary assessments is the backbone), the published papers those assessments cite, read only from copies the user is entitled to, and pinned snapshots of the Materials Project and OQMD.
Code: catalog.json · import_sgte.py · fetch_mp_hull_snapshot.py · stage notes
Each PDF is inventoried once: every table, figure, caption and reference list with its position on the page, and whether it is born-digital or scanned. Every table is then sent to exactly one reading route, or skipped with a stated reason.
Code: inventory.py · routing.py · table_routing.py · stage notes
Deterministic extractors first: table finders, the vector objects of born-digital figures, calibrated tracing of scanned diagrams. Machine readers then read tables and sentences, and must copy the sentence or table row that prints each value. Two independent readers must agree; a judge settles what they do not, and the rest goes to people.
Code: extraction.py · vector_manifest.py · consensus.py · stage notes
Rules any reading must pass: the quote is found word for word on its page and contains the number; every field (phase, method, citation, temperature) is printed in that quote or right next to it; units become K, mole fraction and kJ per mole of atoms only when the basis is printed; lost minus signs are checked on the page image.
Code: quote_fixup.py · flash_window.py · state_requirements.py · stage notes
Every value is experimental, model (tied to the exact database file by checksum), DFT, or unspecified; expert-drawn handbook diagrams are a class of their own. Classes are never mixed or averaged.
Code: origin_rules.py · model_evidence.py · assessed-phase-diagrams/ · stage notes
Binary phase diagrams calculated with pycalphad from every SGTE file, convex hulls from DFT and CALPHAD, and the comparison of literature formation enthalpies with DFT, all from pinned inputs.
Code: compute_binary_diagrams.py · build_binary_hulls.py · dft_check.py · stage notes
Versioned files (JSON, CSV, Parquet), plain pages per system and compound, an MCP server for AI assistants, and Zenodo records with DOIs.
Code: export_literature_data.py · build_parquet.py · orchid-mcp · stage notes
A benchmark of synthetic papers with exact answers, a cross-check against the Kim et al. (2017) compilation of measured formation enthalpies, the DFT comparison, and a hand-checked sample of released values (in progress).
Code: orchid-bench · kim2017_crosscheck.py · build_precision_sample.py · stage notes
The synthetic demo runs the deterministic extractors on two synthetic papers (generated from open SGTE databases, CC BY 4.0), turns the readings into ORCHID records, checks them and scores them against exact answers. It needs no papers, no API keys and no network.
git clone https://github.com/orchid-calphad/orchid-pipeline
cd orchid-pipeline
pip install -e '.[pdf,demo]'
python examples/synthetic-demo/run_demo.py --out demo-outIt prints a score table and writes demo-out/report.html. Step by step: notebooks/synthetic_demo.ipynb. Install options and how
to run the tools on your own papers: the quickstart.
Twenty synthetic papers with exact truth keys, each rendered born-digital, as a clean scan, a degraded scan and an image-only PDF, with the scorer and baselines.
github.com/orchid-calphad/orchid-bench (publication pending)
Lets AI assistants query ORCHID data: systems, compounds, literature values with their quotes, phase diagrams and hulls.
Release v2026.10: ORCHID-Values (literature values), ORCHID-Diagrams (calculated phase diagrams), ORCHID-Hulls, ORCHID-Parity (DFT vs experiment), ORCHID-Registry (phase registry), ORCHID-Assemblages (assessed phase assemblages), ORCHID ESPEI datasets (8,589 datasets) and ORCHID-Bibliography (9,570 papers), as JSON, CSV and Parquet.
A paper describing ORCHID's method and its validation is in preparation; the preprint link will appear here.
Preprint: forthcoming
If you use ORCHID data, cite the dataset you used, the umbrella record, and the original sources of the data. Release v2026.10 (Version 1) is archived on Zenodo as eight datasets, each with its own DOI.
Also cite the original source of every value you use (paper_citation / paper_doi, and cited_dois for values the paper cites from earlier work); B. Hallstedt, The SGTE collection of binary datasets, Calphad 89 (2025) 102833, https://doi.org/10.1016/j.calphad.2025.102833 (CC BY 4.0), and the original assessment named in each file; A. Jain et al., The Materials Project: A materials genome approach to accelerating materials innovation, APL Materials 1 (2013) 011002, https://doi.org/10.1063/1.4812323 (CC BY 4.0); J. E. Saal et al., JOM 65 (2013) 1501, https://doi.org/10.1007/s11837-013-0755-4 and S. Kirklin et al., npj Comput. Mater. 1 (2015) 15010, https://doi.org/10.1038/npjcompumats.2015.10 (CC BY 4.0); G. Kim, S. V. Meschel, P. Nash, W. Chen, Experimental formation enthalpies for intermetallic phases and other inorganic compounds, Scientific Data 4 (2017) 170162, https://doi.org/10.1038/sdata.2017.162; for every diagram used, its original evaluation (citations[role=original_evaluation], DOI where matched) and H. Okamoto, Desk Handbook: Phase Diagrams for Binary Alloys, 2nd ed., ASM International (2010) as the compilation; the original source of every value you use (each dataset's `reference` / `bibtex`, references.bib; index.csv lists the DOIs); the papers themselves when you use their data (doi / citation of each bibliography record).
@misc{orchid_literature_2026,
author = {Arr{\'o}yave, Raymundo and Saal, James and Shin, Dongwon},
title = {{ORCHID-Values: thermodynamic and phase-equilibrium values read from the literature}},
version = {2026.10},
publisher = {Zenodo},
year = {2026},
doi = {10.5281/zenodo.23735569},
url = {https://doi.org/10.5281/zenodo.23735569},
note = {Version 1, machine-extracted}
}
@misc{orchid_phase_diagrams_2026,
author = {Arr{\'o}yave, Raymundo and Saal, James and Shin, Dongwon},
title = {{ORCHID-Diagrams: calculated binary phase diagrams}},
version = {2026.10},
publisher = {Zenodo},
year = {2026},
doi = {10.5281/zenodo.23735859},
url = {https://doi.org/10.5281/zenodo.23735859}
}
@misc{orchid_hulls_2026,
author = {Arr{\'o}yave, Raymundo and Saal, James and Shin, Dongwon},
title = {{ORCHID-Hulls: binary convex hulls (Materials Project, OQMD, assessed CALPHAD)}},
version = {2026.10},
publisher = {Zenodo},
year = {2026},
doi = {10.5281/zenodo.23735891},
url = {https://doi.org/10.5281/zenodo.23735891}
}
@misc{orchid_dft_benchmark_2026,
author = {Arr{\'o}yave, Raymundo and Saal, James and Shin, Dongwon},
title = {{ORCHID-Parity: DFT-vs-experiment formation-enthalpy benchmark}},
version = {2026.10},
publisher = {Zenodo},
year = {2026},
doi = {10.5281/zenodo.23735943},
url = {https://doi.org/10.5281/zenodo.23735943},
note = {Version 1, machine-extracted}
}
@misc{orchid_phase_registry_2026,
author = {Arr{\'o}yave, Raymundo and Saal, James and Shin, Dongwon},
title = {{ORCHID-Registry: phase registry of binary systems}},
version = {2026.10},
publisher = {Zenodo},
year = {2026},
doi = {10.5281/zenodo.23735970},
url = {https://doi.org/10.5281/zenodo.23735970},
note = {Version 1, machine-extracted}
}
@misc{orchid_assessed_assemblages_2026,
author = {Arr{\'o}yave, Raymundo and Saal, James and Shin, Dongwon},
title = {{ORCHID-Assemblages: assessed phase assemblages of binary systems}},
version = {2026.10},
publisher = {Zenodo},
year = {2026},
doi = {10.5281/zenodo.23735996},
url = {https://doi.org/10.5281/zenodo.23735996},
note = {Version 1, machine-extracted}
}
@misc{orchid_espei_2026,
author = {Arr{\'o}yave, Raymundo and Saal, James and Shin, Dongwon},
title = {{ORCHID ESPEI datasets: literature and database-derived thermodynamic fitting inputs}},
version = {2026.10},
publisher = {Zenodo},
year = {2026},
doi = {10.5281/zenodo.23735788},
url = {https://doi.org/10.5281/zenodo.23735788},
note = {Version 1, machine-extracted}
}
@misc{orchid_bibliography_2026,
author = {Arr{\'o}yave, Raymundo and Saal, James and Shin, Dongwon},
title = {{ORCHID-Bibliography: papers and typed citation relations}},
version = {2026.10},
publisher = {Zenodo},
year = {2026},
doi = {10.5281/zenodo.23736021},
url = {https://doi.org/10.5281/zenodo.23736021},
note = {Version 1, machine-extracted}
}
@misc{orchid_data_2026,
author = {Arr{\'o}yave, Raymundo and Saal, James and Shin, Dongwon},
title = {{ORCHID data: umbrella record of the ORCHID datasets}},
version = {2026.10},
publisher = {Zenodo},
year = {2026},
doi = {10.5281/zenodo.23736065},
url = {https://doi.org/10.5281/zenodo.23736065},
note = {Version 1, machine-extracted}
}
Software: cite the code repository as described in its CITATION.cff.