[akoi]

Problem and change

LoopFollow 7.1.0 Statistics can freeze when Download Treatments is enabled and the selected period crosses a daylight-saving fall-back day, such as 25 October 2026 in Europe/Stockholm. Both the scheduled-basal gap loop and temporary-basal adjustment loop advanced by a fixed 86,400 seconds, which can land on the same 25-hour calendar day indefinitely.

Backport upstream fix #795 through a shared calendar-day helper and add three regression tests for autumn, spring and a seven-day range crossing autumn. Retain the three icon-metadata removals from superseded PR #1 for local Xcode 27.1 compatibility. The final diff touches five files; icon images, signing identifiers and upstream version configuration are unchanged.

Branch workflow

This PR is fix/stats-dst-fallback-2026 → release. The new release base starts at official upstream v7.1.0, 4a74b781ad6836dd77eccc03bb08a42749dd59a6. The fork’s main remains an exact upstream mirror. Keeping the existing fix/* head and changing the base follows the documented t1d-infra policy; this PR does not merge upstream dev or place local patches on main.

Validation

• Reviewed source head: 22040408ea85d67a2a2d8dff5b6f12e9f29a32c8.
• Full iOS test target passed with Xcode 27.1 on the iPhone 17 / iOS 27.0 simulator: 100 passed, 0 failed, 0 skipped.
• All three StatsCalendarDayTests passed, including a separate targeted run. These ran in the actual iOS app/test target, replacing the earlier isolated Linux-only validation.
• Compared both loop changes with upstream Fix Statistics hang on daylight saving fall-back days loopandlearn/LoopFollow#795: calendar-day advancement and failure handling match; the helper shares that logic between callers.
• Icon JSON and git diff --check pass. The icon workaround’s earlier controlled actool comparison reproduced the missing renditions with stock metadata and verified generation after removing only the three scale fields.
• Signed distribution archiving, TestFlight upload and physical-device Statistics behaviour remain pending. Simulator success does not establish those outcomes.

Before rollout

• Owner review, then merge this fix into release.
• Update the local checkout to the reviewed release commit; preserve ignored signing configuration, existing bundle identifiers and app groups.
• Follow the LoopFollow-specific guide in akoi/t1d-infra/docs/loopfollow/LOCAL-INSTALL.md and the shared signing/export procedure. Choose an unused build number; the installed/TestFlight baseline is 7.1.0 (2). Prepare versions with agvtool and verify matching app/extension marketing versions and build numbers.
• Create a signed archive, verify required icons and capabilities, and require production push in the exported TestFlight IPA. Check archive entitlements against the archive’s own signing profile.
• Upload to the existing LoopFollow Maya TestFlight record, wait for Apple processing, then install over the current app without deleting it.
• On the follower iPhone, verify Nightscout freshness, treatment/prediction display and Statistics with Download Treatments ON over a range including 25 October 2026. Check period changes and background/foreground responsiveness.
• Tag the distributed source revision and record source SHA, packaged build number and validation outcome in the t1d-infra current setup.

No controller, treatment-parameter, Nightscout-server, credential, Dexcom Follow alarm or remote-command changes. No TestFlight archive or upload was performed for this PR.