Local-first manuscript editing and CRDTs — what user experience should Yjs authority actually enable?

[qnbs]

Why this exists

Issue #555 explores manuscript and Y.Text authority; #548 is a current compatibility defect; and #349 is the collaboration invite-link UX. This Discussion asks what user experience local-first manuscript editing and CRDT authority should actually enable.

Current state

Local-first behavior can provide offline editing, fast local response, recovery, and multi-view continuity for solo authors. Collaboration adds conflict semantics and shared state, but the product should explain what CRDT authority improves even when a user is not collaborating.

The scopes are intentionally distinct:

• local-first: migrate manuscript text to Y.Text and gate the source-of-truth flip #555 = manuscript/Y.Text authority;
• local-first: preserve prototype-named entity IDs when rebuilding ProjectData from Yjs #548 = current compatibility defect;
• Collaboration invite links with expiry/permissions #349 = collaboration invite UX.

Product decision space

Explore solo-author value, offline behavior, undo/redo, multiple views, conflict semantics, collaboration handoff, history/versioning, recovery, and whether CRDT authority earns its complexity without active collaboration. A good model should make acknowledged edits durable and understandable rather than exposing implementation details as new uncertainty.

Role perspectives

Role-perspective note: The viewpoints below are maintainer-curated, AI-assisted design lenses. They are not separate community members, votes, or evidence of consensus.

• ✍️ Solo author: immediate editing, offline continuity, undo/redo, and trustworthy recovery.
• 🤝 Co-author: understandable presence, conflict outcomes, handoff, and collaboration continuity.
• 🧪 Data integrity: durable acknowledged edits, deterministic history, no-loss merges, and explicit recovery.
• ⚡ Large-manuscript performance: bounded memory, responsive views, indexing, and predictable synchronization costs.
• 🧭 Product: use CRDT authority where it improves the writing experience, not merely because collaboration exists.

Questions for the community

• What solo-author problems should Y.Text/local-first solve first?
• How should undo/redo, snapshots, branches, and recovery interact?
• Which conflict outcomes must be visible and user-controllable?
• When does CRDT authority add value without active collaboration?
• How should local-first: migrate manuscript text to Y.Text and gate the source-of-truth flip #555, local-first: preserve prototype-named entity IDs when rebuilding ProjectData from Yjs #548, and Collaboration invite links with expiry/permissions #349 remain separate in user-facing expectations?

Relationship to implementation

This Discussion links product exploration to #555 while explicitly preserving #548 as a technical defect tracker and #349 as the invite UX tracker. Those Issues remain authoritative for implementation, acceptance, priority, and sequencing; the Discussion changes none of them.