I find it interesting how the argument of the current top comment[1] - that denormalized tables are great, as long as you never have any updates for existing entities - basically follows directly from your point 2) :
If for any individual use case a denormalized table fit specifically to that usecase is superior to a normalized table, the most straightforward solution for multiple use cases would be to just keep multiple tables and replicate any insert on each of them. This would work relatively well until the moment where you have to update something...
If for any individual use case a denormalized table fit specifically to that usecase is superior to a normalized table, the most straightforward solution for multiple use cases would be to just keep multiple tables and replicate any insert on each of them. This would work relatively well until the moment where you have to update something...
[1] https://news.ycombinator.com/item?id=35927226