Skip to content

IINTS-AF SDK v1.5.29

Release date: 2026-07-15

Summary

This patch applies a privacy-safe, aggregate OhioT1DM physiology calibration pass to the default free-living reference profile.

The full local OhioT1DM processed splits were audited with iints.research.physiology_calibration to estimate empirical glucose shape features: mean glucose, time-in-range, rate-of-change tails, post-meal peak timing, and dawn summary. The raw train hints suggested a higher basal target and much longer carbohydrate absorption. The basal target was adopted, while carbohydrate duration was constrained by the SDK realism regression gate.

Calibration Inputs

  • ohio_train.csv: 153,055 valid rows, 12 subjects, mean glucose 159.174 mg/dL, TIR 63.916%, median meal peak 110.0 min.
  • ohio_test.csv: 35,925 valid rows, 12 subjects, mean glucose 161.584 mg/dL, TIR 63.468%, median meal peak 97.5 min.
  • ohio_all.csv: 188,980 valid rows, 12 subjects, mean glucose 159.632 mg/dL, TIR 63.831%, median meal peak 110.0 min.

Changed

  • Updated reference_free_living_t1d basal glucose target from 138.0 to 147.0 mg/dL.
  • Kept carbohydrate absorption gate-compatible at 270.0 min after realism regression testing.
  • Documented the calibration result and interpretation limits in the physiology reference.

Verified

  • python3 -m pytest tests/research/test_physiology_calibration.py -q
  • python3 -m pytest tests/test_presets_realism.py tests/core/test_patient_physiology_regression.py tests/core/test_hovorka_model.py tests/data/test_realism_validator.py -q
  • python3 -m pytest tests/ -q
  • python3 -m mypy src/iints/

Safety

This is a research calibration audit, not a patient-specific clinical digital twin. Raw OhioT1DM data remains outside git. No output is intended for diagnosis, insulin dosing, glucagon dosing, treatment decisions, or real-time patient care.