Offline Sync Conflict Resolution

offline-sync-conflict-resolution

Registry Trigger

Use when

• An installed or browser client must read, write, queue, reconcile, delete, or resume work across loss and restoration of connectivity.

Do not use when

• The client is online-only, no synchronization behavior changes, or the open decision is a backend-wide transaction or distributed-consistency policy.

Skill Role

Define client-side authority, pending-operation durability, optimistic reconciliation, revision and tombstone handling, partial synchronization, cursor recovery, unknown results, resumable transfers, and user recovery. Consume generic idempotency and transaction contracts without becoming their owner.

High-Value Rules

• Choose authority per data and operation. Declare local-first, server-authoritative, or explicitly merged behavior before selecting queues or caches.
• Persist pending intent with reconciliation identity. Record the business operation, target identity, base revision, payload version, and user-visible status needed after restart.
• Resolve unknown results before replay. Query authoritative status or use proven duplicate suppression whenever a timeout or disconnect can hide a committed effect.
• Separate optimistic presentation from confirmed truth. Represent pending, accepted, rejected, conflicted, and abandoned outcomes so rollback cannot erase unrelated changes.
• Use semantic revisions instead of client wall clocks. Detect conflicts from authoritative versions or domain intent unless the contract explicitly tolerates clock skew and last-writer behavior.
• Advance cursors only with applied state. Commit page results, deletion markers, and checkpoint progress together so partial synchronization cannot skip or resurrect records.
• Bind resumable transfer state to one upload. Verify resource identity, processed offset, representation length, response completeness, current limits, expiry, and cancellation before appending bytes.
• End blocked work in an owned recovery state. Expose retry, discard, replace, merge, or support escalation according to consequence and user authority.

Anti-Patterns

• Replay every queued request after reconnect without operation identity or authoritative status.
• Drop tombstones before every replica and offline client has crossed the deletion horizon.
• Resolve conflicts by device time while clock skew, account changes, or field-level intent remains material.

Stop Conditions

Stop when authority, business operation identity, conflict policy, deletion horizon, or unknown-result resolution is absent. Escalate shared backend invariants to transaction, concurrency, or distributed-system owners.

Output Contract

• offline-sync decision with authority pending-operation record retry unknown-result behavior optimistic states revision deletion policy partial-sync checkpoint transfer recovery user resolution evidence and proof limits

Targeted References