The 2015 Nepal earthquake revealed that offline-first is not a simple feature but a fundamental distributed systems challenge.
The traditional server-as-truth model fails in partitioned networks, creating divergent histories and unresolvable data conflicts.
Physical device clocks are untrustworthy for ordering events, so logical clocks like HLCs are necessary for correct conflict resolution.
An event-sourcing architecture with a local operation log preserves history and enables intelligent merging, treating the server as a peer.
Avoid silent data loss from "last write wins" by using semantic merge rules and CRDT-inspired types to preserve user intent.
For disaster response, the CAP theorem dictates choosing availability and partition tolerance (AP) to ensure the system works when needed most.
Resilience requires designing for actual field failures like storage exhaustion, device sharing, and social synchronization patterns, not just network loss.
Offline-first architecture is powerful but introduces significant engineering complexity, making it a deliberate trade-off, not a default choice.
Improve system resilience by auditing connectivity assumptions, defining conflict semantics, measuring replication lag, and practicing deliberate failure.
Jobs that call for the skills explored in this talk.