Why CRDT-Based Collaborative Editing Wins for Live Coding Interviews
Two approaches to real-time collaboration
When two people edit the same file simultaneously, the editor has to merge their changes without losing data or causing weird flicker. Two main approaches exist:
- Operational Transform (OT). Each edit is sent as an operation; a central server transforms incoming ops to apply them in the right order. Used by Google Docs.
- Conflict-free Replicated Data Types (CRDTs). Each edit carries enough metadata (typically a unique ID and a logical clock) that any two clients converge to the same state regardless of arrival order. Used by Figma, Linear, and most modern collab tools.
For interviews specifically, CRDTs win.
Why CRDTs win for interviews
Interviews happen on hotel Wi-Fi, mobile hotspots, and corporate networks with aggressive proxies. Three things go wrong:
- Brief disconnects. A 4-second network blip mid-typing.
- Asymmetric latency. Candidate's keystrokes reach the server in 80ms; interviewer's in 400ms.
- Reconnect chaos. Both sides come back online with diverged state.
OT-based editors require the server to mediate every operation. When the network glitches, edits queue up, and reconnects often produce visible flicker, lost characters, or — worst case — silent state loss. Candidates get rattled, interviewers blame the candidate.
CRDTs handle all three scenarios silently. Every client can apply local edits immediately; merges happen on reconnect with no user-visible impact. The editor feels offline-first.
Why Yjs + Monaco became the standard
Yjs is the most battle-tested CRDT library for text. It pairs cleanly with Monaco — the editor that powers VS Code — through y-monaco. The combination gives you:
- VS Code's full editor experience (IntelliSense, multi-cursor, all the keybindings)
- Real-time collab that works on bad networks
- Multiple cursors, selection sharing, awareness presence
- Battle-tested at scale
ClarityHire is built on Monaco + Yjs for exactly these reasons. The candidate's code edits arrive at the interviewer in under 50ms on a healthy connection, and survive network blips that would tank a server-mediated editor.
What this means in practice
If you are evaluating a live coding tool, simulate a bad network during your trial. Throttle to 3G in DevTools, drop the connection for 5 seconds, watch what happens. CRDT-based tools recover invisibly. Others do not.