Star schema or one big table?

Star schema when the model is consumed by multiple teams and tools. Conformed dimensions give consistent filtering and shared definitions, BI tools are built to understand it, and adding a new fact table does not require reshaping anything.

One big table when there is a single workload, the query pattern is stable, and join cost genuinely dominates. Feature tables for machine learning are the clearest case — one row per entity per time, fully denormalised.

In practice most mature platforms have both: a star-schema layer as the governed semantic model, and derived wide tables materialised for specific heavy consumers.

The question is really about who consumes the model. A candidate who answers purely on query performance has missed that a semantic layer's main job is shared meaning, not speed.

The trap answer is "one big table is faster so always use it". A follow-up — "how do you keep the definition of active customer consistent across eleven wide tables?" — usually reveals whether the candidate has operated a platform or only queried one.