Changelog¶
The full changelog lives at the repository root in
CHANGELOG.md
and is reproduced below.
Changelog¶
All notable changes to this project will be documented in this file.
The format is based on Keep a Changelog 1.1.0, and this project adheres to Semantic Versioning 2.0.0.
Unreleased¶
(no changes yet)
0.13.0 - 2026-05-22¶
Added¶
CITATION.cffso GitHub displays a "Cite this repository" button and so third-party citation tools (Zotero, Mendeley) can parse machine-readable citation metadata..zenodo.jsonwith explicit metadata for Zenodo deposit at release time (title, description, author with ORCID, license, keywords).CHANGELOG.mdin Keep a Changelog format, back-filled from v0.7.1.- README badge bar covering CI, license, supported Python versions, current version, and DOI (DOI placeholder until first Zenodo deposit).
- README citation block with both BibTeX and the GitHub-citation button approach.
Changed¶
- Version strings in
pyproject.tomlandsrc/__init__.pysynchronised to0.13.0. Earlier release-bump shell commands used regex patterns anchored on the prior version string; after0.8.0the pattern stopped matching and subsequent bumps silently no-ops'd, leaving the package version at0.8.0while git tags advanced tov0.12.0. The newscripts/update_version.pyreads from any prior version and writes the target explicitly so this cannot recur.
Fixed¶
- Stale
scripts/paper/directory from an earlier manual copy step; the canonical figures live inpaper/figures/.
0.12.0 - 2026-05-22¶
Added¶
- Per-item reporting-standards checklists in
paper/checklists/: tripod_ai_checklist.mdfor the Section 4.6 downstream classifier (TRIPOD+AI, Collins et al. 2024).stard_2015_checklist.mdfor Sections 4.1-4.3 SQI-as-diagnostic-test framing (STARD 2015, Bossuyt et al. 2015).consort_ai_applicability.mdanddecide_ai_applicability.mddocumenting why CONSORT-AI and DECIDE-AI do not apply (not a trial, not a clinical deployment), and what a future deployment study using this toolkit would need to report.- Figure 1: STARD-style data flow diagram (
paper/figures/fig_flow_diagram.pngand.svg), reproducible viascripts/generate_flow_diagram.py. - New manuscript subsection 3.7 "Reporting standards and reproducibility" inserted before Section 4.
Changed¶
- Manuscript text patched to cite the four reporting-standard references
that were previously orphaned in the bibliography
(
@collins2024tripodai,@bossuyt2015stard,@liu2020consortai,@vasey2022decideai).
0.11.0 - 2026-05-22¶
Added¶
- 21 verified bibliography entries appended to
paper/paper.bib: - PPG and wearable validation: Allen 2007, Mejia-Mejia et al. 2021, Schaefer and Vagedes 2013, Bent et al. 2020, Shcherbina et al. 2017.
- Algorithmic fairness in health: Obermeyer et al. 2019, Sjoding et al. 2020, Vyas et al. 2020.
- HRV methodology: Shaffer and Ginsberg 2017.
- Reporting standards: Collins et al. 2024 (TRIPOD+AI), Liu et al. 2020 (CONSORT-AI), Vasey et al. 2022 (DECIDE-AI), Bossuyt et al. 2015 (STARD).
- Causal inference: VanderWeele and Ding 2017 (E-value), Hernan and Robins 2020, Funk et al. 2011 (AIPW).
- Statistical methods: Cohen 1988, Cliff 1993, Wilcoxon 1945, Koo and Li 2016, Adams and MacKay 2007 (BOCPD).
- 13 citation insertions at strategic anchor points throughout the manuscript.
Changed¶
- Total bibliography expanded from 13 to 34 verified references.
Fixed¶
- Removed a single duplicate entry:
taskforce1996(Circulation venue) was a re-print of the existingmalik1996entry (European Heart Journal venue) for the same HRV Task Force 1996 document.
0.10.0 - 2026-05-22¶
Added¶
- Elgendi 2016 (SSQI, KSQI, ESQI) as a third published SQI baseline
alongside Orphanidou 2015 and Sukor 2011. Implementation in
src/signals/elgendi_sqi.pywith auto-polarity detection for the E4 wrist PPG signal convention. - 16 new tests covering
elgendi_sqi(total tests now 235). - Four-way SQI agreement analysis in
src/evaluation/deep_real_analysis.py. - Section 4.2 (manuscript) rewritten around the consensus-rejection finding.
Changed¶
- Section 5.1 Finding #5 generalised from two baselines to three.
- Abstract Results paragraph updated to reflect the three-baseline agreement finding.
- Figure 5 regenerated with four bars and a 4x4 pairwise kappa matrix.
Key result on n=15 WESAD¶
- All three published SQI baselines individually pass 21-26% of wrist PPG windows.
- Pairwise published-vs-published Cohen's kappas: Orph-Sukor +0.41, Orph-Elgendi -0.20, Sukor-Elgendi -0.22 (median +/-0.20).
- All three published baselines simultaneously fail on 44.6% of windows (2,936 of 6,585).
- The in-house default threshold (0.70) passes every one of those 2,936 windows.
0.9.0 - 2026-05-22¶
Added¶
- Section 4.6 downstream-audit demo: per-subject LF/HF ratio (baseline vs stress) classifier evaluated under raw vs in-house cleaned preprocessing.
- LOSO 15-fold cross-validation.
- Paired Wilcoxon signed-rank test on per-fold AUROC.
Key result on n=15 WESAD¶
- Per-subject biomarker correlation rho = +0.10 paired (13 of 15 subjects); Wilcoxon p = 1.5e-4.
- LOSO AUROC 0.804 (raw) -> 0.823 (cleaned); paired comparison p = 0.052.
0.8.1 - 2026-05-21¶
Added¶
- Supplement_extended_analyses.md completed for n=15.
Changed¶
- All figures regenerated at n=15.
0.8.0 - 2026-05-21¶
Added¶
- Real-data validation scaled from n=2 pilot to all 15 WESAD subjects (S2-S11, S13-S17).
Changed¶
- 6,585 windows in total across the full cohort.
Key result¶
- Bland-Altman bias +3.57 bpm between wrist Empatica E4 PPG and chest RespiBAN ECG, 95% LoA [-23.14, +30.28] bpm, MAE 9.66 bpm, Pearson r = +0.70.
- The pilot finding that recalibration improved holdout agreement (delta kappa = +0.22 at n=2) DID NOT replicate at n=15 (delta kappa = 0.00).
0.7.1 - 2026-05-20¶
Added¶
- Initial public release.
- n=2 WESAD pilot validation.
- Synthetic cohort with injected fairness disparities.
- AIPW causal sensitivity analysis.
- Two published SQI baselines: Orphanidou 2015, Sukor 2011.