Consider: Heap-style encoding #369
Description
This allows individual data from a serialized blob to be isolated and read easily. It may be combined with a convention on Merklisation in order to allow for individual fields (or their internal data/lengths if variable size) to be proven.
Rather than encoding in a single pass, have a multi-pass encoder where variable-length items are serialized in two fixed-length quantities as item-count and offset.
While fixed-length, both items are expressed according to the schema.
This comes with some cost to the size, since offsets must be encoded and lengths are not compact. In general both would be 32-bit.