Thanks for posting your thoughts. I'm having a bit of difficulty breaking down "semantic layer" into concrete technological concepts. An attempt:

* Schema definition. PK/FK columns and their relationships. Standard ER fare.

* Dimensions. Partially overlap with PK/FK structure, but may include fields that don't map to an explicit key column, e.g. [trunc] date or zip code or even binned measures. Can see the value of having dimensions documented across a team.

* Measures. Mainly named aggregations, e.g. value = sum(price * quant), which can be aggregated over many dimensional combinations. Definitely useful, though I'd expect PRQL "functions" to be usable in the same role.

* Formatting rules.

Am I missing something crucial?