Upgrading tools would frequently break compatibility with existing data. And it was difficult to make different tools interoperate if they couldn’t agree exactly on the format of their underlying JSON data.

Testing frameworks often introduce their own abstractions for e.g. evaluation order, data validation, reporting, scope, code reuse, state, and lifecycle. In my experience, these abstractions are always needlessly different from (and inferior to) related abstractions provided by the language itself.

As Sebastian Markbage pointed out, no abstraction is superior to wrong abstractions. We are providing low-level components to maximize composition capabilities.

It's much easier to recover from no abstraction than the wrong abstraction.