See the discussion in zulip chat. The implementation relies on the idea described in this message. The implementation in practice linearly scales alpha in an initial section (in practice until delta is 90), at which point the slop of alpha*delta is 0.1. This slope is then continued on in the second section.
The calibration done was based on alpha=0.75 at delta=50; this was estimated to be approximately an hour of withdrawn basal. Note that with an alpha=0.75, this means the result would have 37.5, or be 12.5 mg/dL lower than otherwise. With auto-bolusing, this would result in giving a bolus which would result in 5 mg/dL less effect, which is quite minimal. As negative IOB increases the resulting alpha gets significantly smaller.
Additional testing is necessary for this PR before merging.
See the discussion in zulip chat. The implementation relies on the idea described in this message. The implementation in practice linearly scales alpha in an initial section (in practice until delta is 90), at which point the slop of alpha*delta is 0.1. This slope is then continued on in the second section.
The calibration done was based on alpha=0.75 at delta=50; this was estimated to be approximately an hour of withdrawn basal. Note that with an alpha=0.75, this means the result would have 37.5, or be 12.5 mg/dL lower than otherwise. With auto-bolusing, this would result in giving a bolus which would result in 5 mg/dL less effect, which is quite minimal. As negative IOB increases the resulting alpha gets significantly smaller.
Additional testing is necessary for this PR before merging.