Encoding JSON vs native

[niquola]

To achieve this goal, this document focuses on representing FHIR resources using natively-encoded ANSI SQL structures, including STRUCT, ARRAY, and primitive types. This allows query engines to perform a wide variety of optimizations while offering the most familiar experience to users analyzing data with SQL-enabled tools.

Can we work on unified logical model, which can be also used for JSON (ProtoBuf, etc) databases?
Because a lot of databases native support JSON and SQL - i think, we should not ignore this in SQL-on-FHIR. Optimisations and effective internal representation is implementation details.