Abstracting everything you use has a (high) cost, and doesn't always achieve what you want. This article repeatedly states things "go no further", but realistically, some dependencies will have worse inconveniences than a bad field name. If database you rely on is relational, has token based pagination, and has eventual consistency, chances are that swapping to a consistent timeseries DB with only limit/offset based pagination will impact code in your *.core.* package.