Tune performance

Measure first. In SwiftUI, the cause of jank is almost never where it feels like it is, and an optimisation applied on a hunch usually costs readability for nothing.

Measure

Field data (MetricKit). For what real users experience, use MetricKit. On iOS 27 the Swift-first MetricManager delivers MetricReport and DiagnosticReport values through an AsyncStream; the original MXMetricManager/MXMetricPayload APIs still work but are no longer recommended for new code. Keep the old subscriber for OS versions before 27. If the app references ScrollHitchTimeMetric or scrollHitchTime(_:), recompile with the iOS 27 SDK and move to HitchTimeMetric — otherwise it crashes at launch on iOS 27. Xcode 27's Organizer also adds an Insights Overview of regressions.

Note the number before and after. "It feels better" is not a result.

The usual causes in this architecture

1. Redundant fetches. onAppear() guards on case .idle for a reason — remove that guard and every tab switch refetches. Check:
   • .task on a row triggering paging more than once (loadMoreIfNeeded guards on isLoadingMore).
   • Search not debounced, so four keystrokes are four requests.
   • Two view models built for the same screen because the view was recreated — @StateObject, not @ObservedObject, for owned view models.
2. Whole-list invalidation. ObservableObject invalidates the view on any @Published change. So:
   • Keep the primary content in one LoadState and avoid publishing properties that change on every keystroke alongside it where possible.
   • Give ForEach stable identity (Identifiable with a real id, never .indices).
   • Keep row views cheap: no date formatter created in body, no filtering or sorting in body. Precompute in the view model.
3. Formatter and encoder churn. DateFormatter and JSONEncoder are expensive to create. This project keeps them as statics (JSONDecoder.api, ISO8601DateFormatter.api). Never construct one inside body or inside a loop.
4. Cache writes on the hot path. FileCacheStore writes whole JSON files. For a large collection, write once per fetch rather than per item, and consider caching only the first page or two — a user who is offline needs the recent screens, not everything.
5. Launch work. AppEnvironment.bootstrap() runs before the first screen. It should do the minimum: restore the session, start the monitor, count pending items. Draining the outbox is fine because it is awaited off the critical path; anything that blocks the first paint is not.
6. Sync frequency. Sync is triggered by connectivity returning, the app becoming active, and explicit refresh — deliberately not a timer. Do not add one; a repeating timer that fires while offline burns battery for nothing.
7. Images. Not covered by this template. If images arrive, add a bounded cache with downsampling before decode (CGImageSourceCreateThumbnailAtIndex) — full resolution decoded into a 60-point row is the most common memory bug in list screens.

Retain cycles to check

• Task { } inside a view model capturing self strongly and outliving it — use [weak self] for anything long-running.
• Closures stored as properties (onSignedOut, setStateHandler) — both use [weak self] at the capture site; keep it that way.
• NWPathMonitor.pathUpdateHandler capturing self — already [weak self].

Verification

• A measurement before and after, with the numbers stated.
• The fix addresses what the measurement showed, not what you guessed.
• No readability sacrificed for an unmeasured gain.
• Request count for the scenario checked in the networking log — no duplicates.
• Memory stable after ten navigations in and out of the screen.
• No new caching layer added without a cap and an eviction story.