Online collaboration is becoming prevalent in our day-today work. As commercial applications show, next to texts, spreadsheets are an essential tool for storing and organizing shared data. However, concurrent modifications of a collaborative spreadsheet can lead to unexpected results when they reflect implementation decisions rather than user intention.

With this paper, we systematically discuss spreadsheet operations and their semantics and propose intention-preserving designs in a concurrent decentralized setting, thus supporting offline operations. We further explore different data models for shared spreadsheets based on composed Conflict-free Replicated Data Types (CRDTs) and give an implementation in the local-first framework Yjs.