v1.5.11¶
v1.5.11 is the AGP/XAI research-readiness release. It packages the latest physiology, explainability, updater, reporting, and local-AI workflow improvements so demo and research installs can update cleanly from GitHub or PyPI.
Highlights¶
1. Publication-grade AGP exports¶
AGP-style reporting now exports both high-resolution PNG and vector SVG assets:
iints report \
--results-csv results/one_day/results.csv \
--style agp \
--png \
--svg \
--bundle-dir results/one_day/agp_report
The AGP asset bundle includes:
agp_report.pdfagp_summary.jsonagp_assets/agp_profile.pngagp_assets/agp_profile.svgagp_assets/daily_profiles.pngagp_assets/daily_profiles.svg
The visual style uses publication-oriented Matplotlib defaults, SciencePlots when installed, clinical time-in-range colors, and a clearer single-day AGP fallback.
2. Explainable AI event stream¶
The simulator now emits explainable_events for clinically relevant moments during run_live, including meal response, faster-than-expected absorption, hypo-risk onset, and supervisor interventions.
AGP asset export now writes:
xai_events.txtfor human-readable demo/review workflowsxai_events.jsonfor downstream indexing, training, and analysis
The XAI engine now uses a dedicated previous-trend state and cooldowns, so explanations are less noisy and more faithful to the actual simulation state.
3. Stronger physiology and safety modeling¶
This release folds in the latest physiology work across patient models and simulator safety paths, including Hovorka-style multi-compartment support, stress/illness effects, glucagon safety rails, HAAF/counter-regulation modeling work, sensor/CGM realism improvements, and more realistic preset behavior.
4. Local AI and results workflows¶
The research tooling now better supports glucose forecasting, local-AI data preparation, results indexing, and privacy-safe OhioT1DM aggregate workflows. Private/raw gated datasets should remain outside git; docs use OHIO_T1DM_ROOT examples rather than local absolute paths.
5. Easier maintenance commands¶
iints update has stronger repair/fallback behavior, and iints delete provides a visible deletion plan for removing SDK packages, user caches, plugins, and optional generated outputs.
Validation¶
Release-focused validation run before tagging:
python3 -m pytest tests/core/test_simulator_xai.py tests/regression/test_pdf_regression.py -q
python3 -m mypy src/iints/core/simulator.py src/iints/analysis/reporting.py tests/core/test_simulator_xai.py tests/regression/test_pdf_regression.py
python3 -m flake8 src/iints/core/simulator.py src/iints/analysis/reporting.py tests/core/test_simulator_xai.py tests/regression/test_pdf_regression.py
python3 -m mkdocs build --strict
Notes:
- The full private OhioT1DM dataset is not included in the repository.
- Generated demo folders such as
deliverables/anddigital_patient_runtime/are not part of the release source package. - IINTS remains research/education software, not a medical device or treatment recommendation system.