A consistency model describes what observations a client may make when operations overlap or replicas lag. Linearizable reads behave as though each operation takes effect at one point between its call and response. That helps with a seat reservation, where two successful claims would be a defect. It costs coordination when the authoritative replicas can't communicate.
Other workflows need a narrower guarantee. A session can require its own writes without requiring every reader to see them immediately. A replicated counter can merge independent increments when the business accepts a delayed total. Start with the operation, then choose the guarantee. Calling a whole application strongly consistent hides these distinctions and makes it hard to explain which work remains available during a partition.