This specification defines how mutation information moves between peers and clients. Mutation semantics are defined in mutations.md, conflict precedence in reconciliation.md, and storage lifecycle in persistence.md.
- Peers receive operation-tagged mutation/update objects with the addressing, original timestamp and identity, and reconciliation metadata needed to apply them.
- Ordinary peer mutations propagate update objects rather than resulting document snapshots.
- Receiving peers reconcile incoming operations against their own state before applying them.
- Duplicate delivery must not apply relative operations more than once.
- A reconciled update must retain enough original ordering information for downstream peers to make the same decision.
Clients receive the resulting document state together with the corresponding mutation/update object. Deletions include explicit deletion state.
Return this response as soon as the in-memory CRUD result is ready. Peer delivery, persistence, and client acknowledgment are not prerequisites. CoCreateDB does not wait for a client acknowledgment before continuing backend work; durable client-side recovery is a separate client responsibility and protocol.
CoCreateDB owns peer message shape and mutation application. Mesh owns routing and delivery mechanics. CRUD Client/Server and CRDT callers may handle their own offline intent, validation, and pre-submission reconciliation, but CoCreateDB still applies its specified reconciliation policy to peer mutations.
Message delivery, retries, durable acknowledgments, and multi-file atomicity are
separate guarantees. A successful send alone does not establish durable replica
acknowledgment. Current implementation gaps and alignment work are tracked in
NEXT_TASK.md.