Session Analytics Are Wrong for Users Who Go Offline

A mobile app logs user interaction events locally when the device is offline and batch-uploads them once connectivity returns, each event carrying its original client-side timestamp. A streaming job computes "session windows" (30-minute inactivity gap) over these events to measure session length and engagement, using processing time (the time the event is received by the server) rather than the client-side event timestamp. Analytics report a large spike in artificially long, low-engagement "sessions" specifically for users who use the app on subways or in areas with poor connectivity.

1. Explain the mechanism by which using processing time produces this specific symptom (artificially long sessions for offline-prone users).
2. Would switching to event time alone fully fix session-length accuracy for this population? What additional consideration does event time introduce that processing time didn't have?
3. Describe the concrete change(s) you'd make, including how you'd validate the fix before rolling it out broadly.