PyTM30 implements ANSI/IES TM-30-20 from the published standard:
ANSI/IES TM-30-20, IES Method for Evaluating Light Source Color Rendition, An American National Standard, Illuminating Engineering Society, New York, 2020, incorporating Errata 1 (2021). ISBN 978-0-87995-379-9. (IES standards are issued without DOIs.)
The spectral data tables in data/ are CIE datasets, published by the
International Commission on Illumination (CIE) under CC BY-SA 4.0 and
distributed here under the same licence, separately from the MIT code.
data/README.md lists each file's source DOI and the changes made. They were
extracted using colour-science (BSD-3-Clause) as a tool:
Mansencal, T., Mauderer, M., Parsons, M., et al. (2025). Colour 0.4.7 [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.17837391 Repository: https://github.com/colour-science/colour Website: https://www.colour-science.org Per-table provenance is recorded in tools/generate_data_colour_science.py. Every shipped table is byte-reproducible by running that script and tools/generate_planckian_lut.py on the generating platform, macOS arm64, against colour-science 0.4.7 (verified byte-identical there under both numpy 2.3.3 and numpy 2.5.2; last verified 2026-08-14, all tables, zero byte differences). Byte-identity does NOT transfer across architectures: numpy and libm round differently in the last ULP (observed max 17 ULP regenerating on Linux x86_64), so the data-reproducibility CI job regenerates the tables on Linux with the same pins and enforces numeric equivalence instead -- identical structure and every value within 64 ULP (or 1e-12 absolutely, for near-zero cancellation results) via tools/check_data_reproducibility.py. Byte-level reproducibility is claimed only for the pinned combination on macOS arm64; other versions of colour-science or numpy have not been tested. The CES reflectance data and the TM-30-20 method originate with the IES and CIE.
Numerical accuracy is cross-validated against colour-science (BSD-3-Clause). The golden test fixtures are produced by tools/generate_fixtures.py from colour-science primitives plus the standard's own equations; see docs/divergences.md for exactly which quantities are independently validated and by what.
PyTM30 was first developed outside this repository, with luxpy (GPL-3.0) as its main accuracy oracle: the C++ implementation and its test suite were validated against fixtures generated with luxpy. Before the first commit here (35848d4), the data tables and fixtures were regenerated from CIE data using colour-science, so that commit already contains a finished implementation.
Until 2026-08-14, some sites followed luxpy's conventions where luxpy departs from TM-30-20: the Rcs/Rhs scaling, the CVG display scale and reference hue angle, and a shifted-triangular CCT blend that was always on. A remediation series merged that day (3505b16) changed those sites to follow the standard, regenerated the fixtures, recomputed the test literals that were still luxpy output values (e49e69f), and replaced luxpy-based justifications in comments with citations to TM-30-20 and its normative references. Most clause citations were attached to existing code in that series: they record which clause each site implements, not the order in which the code was written.
Where the standard is silent, two choices still follow luxpy's TM-30 configuration: the extent of the Planckian LUT (tools/generate_planckian_lut.py) and the optional shifted-triangular blend, now off by default. The Python layer's result names follow luxpy's spd_to_tm30() for compatibility. Other implementation choices the standard does not dictate are listed in docs/divergences.md.
luxpy is still used from time to time as a comparison oracle during development. It is not a dependency, and no luxpy code is present in this repository or its history.
Where TM-30-20 and existing implementations differ, this library follows the standard. See docs/divergences.md.