Peritext: A CRDT for Rich-Text Collaboration

Real-time collaboration tools like Google Docs can feel stressful and performative for professional writers. Interviews with eight professionals revealed preferences for private editing spaces with the ability to merge work when ready, analogous to Git's branching model but for rich text documents.

The paper develops a comprehensive model of user intent through nine detailed examples covering concurrent insertions, formatting overlaps, conflicting marks, span boundary behavior, and comment handling. This model serves as a test suite for evaluating whether a merge algorithm produces correct results.

Encoding rich text as Markdown in a plain-text CRDT fails because concurrent formatting operations produce interleaved control characters that change semantic meaning. For example, two users independently bolding overlapping regions produces incorrectly nested asterisks that unbold text that should remain bold.

Using start/end control characters (like HTML tags) in a plain-text CRDT also fails because accumulated state doesn't contain enough information to handle overlapping spans correctly. Nested bold tags cause premature span termination, and the approach lacks temporal ordering information needed to resolve conflicts.