Offline Sync Anomalies

They are also a realistic example of a deployment that cannot rely on a globally synchronized server clock to keep operating usefully.

The Failure Mode

A mobile device can create events while disconnected:

1. draft_created
2. draft_edited
3. draft_submitted

Hours later, the device reconnects and uploads them after server-side events have already happened:

• review_started
• notification_sent

Naive sorting by server ingestion time can make the offline device appear late to its own history.

Why This Happens

Offline systems break the hidden assumption that creation time, observation time, and ingestion time are close enough to be interchangeable.

That gap is not always a few seconds. In real deployments, a central server may be down for 4 to 8 hours while devices or nodes continue locally and sync only when the server is available again.

Those timelines are not interchangeable.

What The Library Should Surface

The library should help a developer distinguish:

• local monotonic sequence on the device
• server ingestion order
• derived HLC order across nodes
• places where order is only partially knowable

This is where same_node_sequence and explicit evidence matter.

Why This Matters

If an offline sync system is forced into one clean server-side order, the application may accidentally tell a false story about user intent and state evolution.

causal-order is useful here because it lets the system preserve ambiguity where ambiguity is real.