Field-level conflict resolution
Proposal, not an available feature. No network permission or new data collection is enabled. These RFCs are locally prepared; remote review/merge awaits the owner repository.
Current architecture
Sync currently compares entries but replaces a complete copy after review. There is no authenticated last-common-base history.
Design under review
Persist only encrypted last-common-base snapshots per trusted peer and validate their revision identity. Match entries/groups by KDBX UUID. Apply a field only when one side changed from the base; present both values when both changed differently. Treat deletion versus modification, moved pages, attachments, unknown fields and schema versions explicitly. Fall back to complete-copy review when the base is absent, untrusted or incompatible. Never combine master-credential changes automatically.
Milestones
- M1: pure merge proposal model with property tests and encrypted-base lifecycle
- M2: conflict UI requiring explicit decisions and atomic encrypted commit/backup
- M3: native two-device race/disconnect/replay tests and KeePassXC round trips
Required acceptance tests
- Nonconflicting edits commute; conflicts never silently drop an edit
- Delete-versus-edit, duplicate UUID and cyclic page moves require review or fail closed
- Failed save leaves the current vault and encrypted conflict recovery copies intact
Limits
No automatic merge is enabled. Base snapshots increase encrypted local retention and require a policy update before implementation.
Any implementation must preserve local unlock/export, add KDBX/KeePassXC regressions for crypto/format changes and update privacy disclosures before new data flows begin.