[ps2]

Depends on LoopKit/LoopKit#683 for the chart's guard line. LoopKit/LoopAlgorithm#16 (Guard + Runway v2 and slow retrospective correction) is already in LoopAlgorithm main; the LoopWorkspace bump, LoopKit/LoopWorkspace#568, brings it in.

Slow retrospective correction

A new row under Settings → Algorithm Experiments, with its own description and toggle, off by default.

• History. The experiment keeps no state, so when it's on fetchData reads:

  • 72 h of glucose instead of 12 h
  • carbs from 72 h plus the longest absorption time (10 h)
  • doses from 72 h plus the insulin activity duration

  Carb ratio, ISF, basal and override history follow those windows, since they are derived from them. That satisfies the algorithm's new carbRatioOrSensitivityStartsAfterCarbEntry check. With the experiment off, nothing changes.

• Recorded. The dosing decision's slowRetrospectiveCorrection experiment carries level and multiplier.

• Matching forecasts. The bolus screen, the manual dose screen and forecast details draw their own forecasts with predictGlucose. It now takes the ISF multiplier from the algorithm's last run, so those forecasts match the algorithm's.

• Forecast details. A Slow Retrospective Correction row shows the ISF multiplier and the level. Deselecting it draws the comparison forecast at the scheduled ISF.

• Analytics. An on/off user property and an experiment-change event, as for the other experiments.

Guard + Runway v2

• Labels. Labeled "Guard + Runway v2" in settings and on forecast details.
• Description. Rewritten for v2: the volatility band lowers only the forecast the low guard checks, not the displayed forecast or the size of corrections, and the runway fades out as carbs rise to 20 g instead of stopping at any carbs.
• Recorded. The dosing decision's guardAndRunway experiment carries version: 2, so v1 and v2 data can be told apart.

The guard on forecast details

While Guard + Runway is on and its row is selected, forecast details draws the forecast the low guard checks as a dotted line. That's GuardAndRunwayOutput.guarded(predictedGlucose), shown when the band applied, and the row says so. Deselecting the row hides it.

The comparison forecast is no longer lowered by the band, because in v2 the displayed forecast isn't.

Testing

• All 426 LoopTests pass, including 2 new ones:
  • testSlowRetrospectiveCorrectionReadsThreeDaysOfHistory: with the experiment on, glucose from 71 h ago and a carb entry from 80 h ago are read, and the carb ratio and ISF timelines reach back past that entry. With it off, neither is read.
  • testSlowRetrospectiveCorrectionIsRecordedOnTheDosingDecision: three days of an unexplained slow rise give a positive level and a multiplier below 1, and the multiplier is recorded on the loop's dosing decision.
• In the iOS Simulator with the CGM simulator:
  • both experiment screens
  • the guard's dotted line, and hiding it with the row
  • the slow RC row