gpbiometricspy 0.1.8 — PPG measurement equity and STEP evidence

Published 2026-10-04.

gpbiometricspy 0.1.8 preserves the frozen 406/406 gpbiometrics 2.0.0 semantic contract with 0 pending exports while promoting the post-0.1.7 PPG pigmentation / measurement-equity programme into the stable release scope.

Read the PPG measurement-equity article Run the synthetic example View the 0.1.8 figures Inspect STEP evidence

Exact release source and qualification

The immutable v0.1.8 tag and public release artifacts resolve to exact stable source:

32cafd5565bdf4a7575dcf5c5d6bb2bfbe101e5d

The scientific qualification baseline immediately before the stable freeze was a75eb1d73791213e6e830ff94975f632455632e2. The merged 0.1.8 source then passed the complete protected exact-main release gate set before the immutable tag and public artifacts were created.

• 12/12 OS × Python 3.11–3.14 test lanes green;
• 874/874 tests;
• 15,612/15,612 statements = 100.00%;
• 7,315/7,334 raw branches = 99.7409%;
• exactly 19 reviewed structural arcs;
• 0 unexpected / 0 stale / 0 unaudited branch debt;
• 7,334/7,334 = 100.0000% audited branch accounting;
• docs/Pages, CodeQL, deep parity, interoperability, private real-data validation and release-handoff safety green;
• Windows install/repair/upgrade/downgrade release-policy proof green;
• canonical GitHub Release and PyPI Trusted Publishing completed successfully.

Public release artifacts

• GitHub Release: v0.1.8
• PyPI: gpbiometricspy 0.1.8
• wheel SHA-256: b1a88c2306843df2cc324b8987335bba413d27715a81f451ba0a036afe0a39f0
• sdist SHA-256: 231c14bab9eeddb0e0e6e9dce0cc095451cf48d7dcd160c5c553005482d7c24d

PyPI Trusted Publishing generated and uploaded digital attestations for both distributions.

PPG measurement-equity surface

Version 0.1.8 includes the Python-native functions compute_skin_ita(), validate_skin_pigmentation_metadata(), summarize_ppg_quality_by_pigmentation(), compare_ppg_reference_by_pigmentation() and ppg_pigmentation_audit().

The method separates pigmentation provenance, acquisition quality, retention/paired availability, and reference agreement. Objective CIELAB/ITA-style measures can support continuous descriptive associations. Subjective Fitzpatrick/Monk-style scales remain subjective provenance; demographic race/ethnicity proxies are not treated as pigmentation.

The package does not apply a universal pigmentation correction, does not automatically exclude observations because of pigmentation, and does not label a device “fair” or “unfair”.

Acquisition quality. Raw waveform/SQI evidence is retained as a separate analytical layer.

Availability. Candidate-HR retention varies across ITA in the known-truth generator.

Reference agreement. HR error remains ITA-neutral in the same generator.

See PPG pigmentation & measurement-equity audit, the synthetic worked example, and the explanation article Auditing PPG measurement equity.

STEP external evidence

BigIdeasLab_STEP v1.0 was executed on an authorized local copy under the applicable PhysioNet DUA against exact scientific source dc2b539154f65061a90d360f01503108ad1fd39f.

The retained aggregate record includes:

• DOI 10.13026/cqfy-d860;
• 53 participants;
• 1,431,570 long-format wearable rows;
• 1,328,850 rows with ECG reference available;
• 361,675 paired wearable/ECG measurements;
• six wearable devices;
• participant-cluster bootstrap with 1,000 replicates and seed 20261003;
• derived-only bundle SHA256 65EF1DE84CF711F9E6753DB771E3E2386FD6926A104E0D4DC1618EE88E8744B1.

Restricted participant-level rows were not committed or redistributed. The visual above uses only DUA-safe aggregate counts already retained in the public evidence record.

Fitzpatrick 1–6 is retained as subjective categorical phototype, not objective pigmentation. STEP row-level retention is interpreted as paired availability/reporting density because device reporting cadence differs materially. The observed unlabeled/NA activity stratum remains unlabeled. Agreement results are observational and device/activity-specific.

See the external-evidence record.

Explicit exclusions

0.1.8 does not claim:

• causal effects of pigmentation or melanin on PPG performance;
• race or ethnicity as a substitute for measured pigmentation;
• generic device fairness from paired error statistics;
• a universal pigmentation correction;
• real ENCoDE execution;
• clinical SpO₂/pulse-oximeter validation.

ENCoDE remains deferred future evidence infrastructure.

Archival state

The software concept DOI remains 10.5281/zenodo.22150872 and the frozen R reference DOI remains 10.5281/zenodo.21434608. A 0.1.8 version DOI is not claimed until genuine independent Zenodo ingestion is verified.

Publication state

gpbiometricspy 0.1.8 is the current public stable release. The immutable v0.1.8 tag, GitHub Release assets, canonical wheel/sdist hashes, PyPI publication through Trusted Publishing, and digital attestations have all been verified against exact source 32cafd5565bdf4a7575dcf5c5d6bb2bfbe101e5d.