Case study: Collaborative editor (Google Docs)

Why must a collaborative editor apply the user's keystroke locally before waiting for the server's acknowledgement?

50 editors in the same document each broadcast 4 cursor movements per second to the other 49 editors. Why must cursor updates never be written to the operation log database?

Why does the Google Docs (Jupiter) protocol allow a client only one unacknowledged operation batch in flight at a time?

A document has accumulated 250,000 operations over six months. How does the system open it in under 100 ms without writing a full 100 KB snapshot on every single keystroke?

Starting from 'CAT', Alice sends Insert(0, 'S') ('SCAT') and Bob concurrently sends Delete(2) (intending to delete 'T'). If Alice's op is applied first, what must OT transform Bob's op into?

What is the main storage cost of sequence CRDTs (such as RGA) compared to central-server OT?

5,000 employees open the company all-hands document at once. How do you prevent $5,000^2$ cursor broadcasts and transformation stalls?

The collaboration server holding the in-memory session for `docId = 42` crashes. How does the replacement server resume without losing acknowledged edits?